digital.library.unt.edu/67531/metadc330688/m2/1/high... · testing of the first cir model involved...
TRANSCRIPT
3.11 A/8/J
h o . A R O 5 ~
FUTURES-FORWARD PRICE DIFFERENCES AND
EFFICIENCY IN THE TREASURY BILL
FUTURES MARKET
DISSERTATION
Presented to the Graduate Council of the
North Texas State University in Partial
Fulfillment of the Requirements
For the Degree of
DOCTOR OF PHILOSOPHY
By
Alan Wong, M.B.A.
Denton, Texas
May, 1986
\ V )
Wong, Alan, Futures-Forward Prioe Differences and
Efficiency in the Treasury Bill Futures Market. Doctor of
Philosophy (Finance), May, 1986, 8 tables, bibliography,
58 titles, 97 pp.
This study addressed two issues. First, it examined
the ability of two models, developed by Cox, Ingersoll and
Ross (CIR), to explain the differences between futures and
implicit forward prices in the thirteen-week T-bill market.
The models imply that if future interest rates are stochastic,
futures and forward prices differ; the structural difference
is due to the daily settlement process required in futures
trading. Second, the study determined the efficiency of the
thirteen-week T-bill futures market using volatility and
regression tests. Volatility tests use variance bounds to
examine whether futures prices are excessively volatile for
the market to be efficient. Regression tests investigate
whether futures prices are unbiased predictors of future
spot prices.
The study was limited to analysis of the first three
futures contracts, using weekly price data as reported m
the Wall Street Journal from March, 1976 to December, 1984.
Testing of the first CIR model involved determination of
whether changes in futures-forward price differences are
related to changes in local covariances between T-bill
p i i \ ) f
futures and bond prices. The same procedure applied in
testing the second model with respect to changes in futures-
forward price differences, local covariances between T-bill
spot and bond prices, and local variances of bond prices.
Volatility tests of market efficiency involved comparison
of mean variances on both sides of two inequality equations.
Regression tests involved determination of whether slope
coefficients are significantly different from zero.
The results provide no evidence that the CIR models
can adequately explain T-bill futures-forward price differ-
ences. Measurement errors could have contributed to the
inconsistent results. It is also possible that the models
are valid only with non—financial underlying assets. The
volatility and regression test results indicate that the
T-bill futures market is efficient. However, the regression
test results show that autocorrelation exists in the nine-
month futures contract data, probably due to a missing
variable related to information costs or default risk.
TABLE OF CONTENTS
Page
LIST OF TABLES v
Chapter
I. INTRODUCTION . . 1
Background and Significance of the Problem
Usefulness of the Study Theoretical Models Outline
II. THE TREASURY BILL FUTURES MARKET 10
Emergence of the Treasury Bill Futures Market
Characteristics of the Treasury Bill Futures Market
Advantages of Forward Trading in Futures Markets
III. REVIEW OF LITERATURE 20
Futures-Forward Price Difference and Transaction Costs
Futures-Forward Price Difference, Daily Settlement and Taxes
Variance Tests of Market Efficiency Summary
IV. METHODOLOGY 46
Hypotheses Procedures Data Summary
V. RESULTS 67
Tests of the Cox-Ingersoll-Ross Propositions
i n
Chapter Page
Intra-Sample Variations Among Futures-Forward Price Differences
Variance Tests of Market Efficiency Regression Tests of Market Efficiency Summary
VI. SUMMARY AND CONCLUSIONS 87
Implications for Further Research Conclusion
BIBLIOGRAPHY 92
IV
LIST OF TABLES
Table Page
I. Data Behavior During Sample Period 59
II. Autocorrelations of T-Bill Spot Price 62
III. Autocorrelations of Spot and Futures Percentage Price Changes 64
IV. Test Results of CIR Models 68
V. Intra-Sample Variations of Futures-Forward
Price Differences 72
VI. Variance Test Results of Equation [20] . . . . 75
VII. Variance Test Results of Equation [23] . . . . 77
VIII. Regression Test Results 80
CHAPTER I
INTRODUCTION
Background and Significance of the Problem
Commodity futures and forward markets have long served
a very important economic function—hedging away risk.
Hedging buyers and sellers can pass the risks they intend
to avoid on to the professional speculator. The fact that
commodity exchanges have been growing for more than a
century indicates the importance of the risk-averting func-
tion. Speculators relieve commodity producers and businesses
of their buying and selling risks so that they can concen-
trate on efficient production. In spite of the attention
futures and foward markets have received, many commodity
traders and economists do not recognize that these two
types of markets are fundamentally different. Although the
two markets look similar, they are not identical. This
study explains the underlying contractual difference between
forward and futures markets; profits or losses from a
futures contract are settled daily between participating
parties while those from a forward contract are settled at
maturity. This paper empirically investigates the differ
ence between T-bill futures and implied forward contracts
using two models—developed by Cox, Ingersoll and Ross (1),
hereafter CIR. CIR propose that stochastic interest rates
contribute to futures-forward price differences due to the
daily settlement requirement in futures trading.
This study also tests the efficiency of the thirteen-
week T-bill futures market. Since the International Monetary
Market of the Chicago Mercantile Exchange first introduced
the T-bill futures contract on January 6, 1976, a number of
researchers have tested the futures market's efficiency.
These efficiency tests dealt mostly with possible arbitrage
opportunities by comparing future yields with the corre-
sponding adjusted forward yields implied in the term struc-
ture of the spot T-bill rates. The implicit forward yields
were adjusted for market imperfections—mainly transaction
costs and taxes. Any significant and sustained differential
between the futures and the forward yields suggests profit-
able arbitrage opportunities and, therefore, market ineffi-
ciency. The assumption of different arbitrage conditions
and the choice of different sample periods are some of the
shortcomings of the past studies. More important, the fail-
ure to subject data of the same period to alternative
techniques impairs the validity of the conclusions. Each
technique has its limitations, and some may not effectively
detect certain complicated futures and forward price move-
ments—other tests might. Since one cannot compare the
power of different techniques, alternative techniques should
be used to determine whether the results are technique-
sensitive. This study examines the efficiency of the
thirteen-week T-bill futures market using regression and
variance tests. Regression tests assume that futures prices
are unbiased predictors of future spot prices. The variance
tests involve formulation of variance bounds to determine
whether T-bill futures prices are too volatile for the market
to be efficient.
This study, thus, addresses the following questions.
1. Are CIR able to explain the differences between
T-bill futures and implicit forward prices?
2. Is the thirteen-week T-bill futures market
efficient, based on the use of regression and
variance tests?
The study finds that the CIR models are unable to explain
the price differences and that the market is efficient.
Usefulness of the Study
Many practitioners apparently believe that futures and
forward contracts are synonymous and regard any discrepancy
between futures and forward prices as evidence of market
inefficiency—the presence of profitable arbitrage oppor-
tunities. If futures and forward markets differ fundamen-
tally, the supposed market inefficiency might lead
arbitrageurs to take up trading positions that would not be
beneficial. Moreover, a trader who wishes to hedge by
taking a long position in the futures market and a short
position in the forward market does not necessarily take a
riskless position. If interest rates are stochastic, such
a forward-futures hedge is risky. This study seeks to
empirically clarify an obscure, but basic, difference
between futures and implied forward contracts in the thirteen-
week T-bill futures market. Some have examined the theoreti-
cal implications of the difference, but empirical tests of
the theory have not yet been published.
The T-bill futures contract is relatively simple to
price due to the uniformity and liquidity of the contract's
underlying asset. Traders make bids and offers on bill
futures contracts by open outcry on the floor of the futures
exchange. Given the ease of pricing and contract and a
highly competitive system, one would expect the T-bill
futures market to be highly efficient. However, past studies
produce conflicting results. An inefficient bill futures
market might impair governmental borrowing capacity. More-
over, T-bill futures data provide valuable help to many. It
provides economic theorists with testable data on the forma-
tion and predictive accuracy of expectations. One weakness
of earlier empirical studies of term structure is the absence
of such observable data. Observable interest rate forecasts
also provide policymakers and forecasters a counter-check on
their econometric forecasts. If the pricing of the T-bill
futures contracts is inefficient—prices do not fully
incorporate available information—they would provide biased
forecasts of subsequently observed spot prices. The useful-
ness of bill futures price data to economic theorists,
policymakers and other interested parties depend on pricing
efficiency.
The efficiency of the T-bill futures market is, there-
fore, important. Since past authors have mostly investi-
gated the issue with arbitrage arguments, this study examines
it with alternative approaches—variance and regression
tests.
Theoretical Models
The regression and variance tests are based on the
efficient market model. This study uses the Cox-Ingersoll-
Ross models to derive tests that can determine whether the
futures-forward price differences are attributable to sto-
chastic interest rates. The CIR model is only briefly dis-
cussed here. It is necessary to leave a more detailed
exposition of CIR's work for the literature review chapter.
Market Efficiency Model
The efficient market model is a fair game model. A
fair game implies that, on average, the actual price of an
asset in the future is equal to its expected price, based on
information at the time. Symbolically,
E{xt+1 " E(xt+l/3:t)} = °' [1]
where E(.) = expectations operator,
xt +l = price at time t+1, and
It = information set at time t.
In other words, expectations are not biased and they result
in zero expected profits.
Efficiency in futures markets is often stated in terms
of a fair game. The relationship between any futures price
and the expectations of its realization is
Ft,n = V W ' n>t 121
where E (.) = expectations operator,
1^ = information set at time t,
Sn = spot price at time n, and
Ft n
= futures price at time t, quoted for delivery
at time n.
An equivalent expression of Eq. [2] is
S = F +U. [31 n t,n t,n 1 J
where U. = S - E (S ). t, n n t n
According to Fama (2), Eq. [2] is an efficient market
model if the forecast error is a fair game. Mathematically
speaking,
Vut,„' = °' W
The forecast error is orthogonal to any element of the
information set; therefore, it is serially uncorrelated.
Eq. [2] is the model for the tests in this study.
Cox-Ingersoil-Ross Models
Proposition 6 in CIR theoretically establishes the
futures-forward price differential in terms of the relation
between futures and bond prices. Assuming a continuous-
time, continuous state economy, the proposition states that
the futures-forward price difference is equal to s
F(t,s) - f (t, s) = PVfc {-£F(U) [covF(u),B(u)]du/B(t)}, [5]
where cov{F(u),B(u)} = local covariance of the percentage
changes in futures and bond prices,
f(t,s) = forward price at time t for a con-
tract that matures at time s,
PVt s = current value at time t of a payoff
amount obtained at time s,
F(t,s) = futures price at time t for a contract
that matures at time s, and
B(t) = price of a default-free discount bond
at time t paying $1 at time s.
Proposition 9 in CIR expresses the difference between
futures and forward prices in terms of spot and bond prices.
Assuming the underlying asset does not make any payout
between t and s in a continuous-time, continuous-state
economy, the futures-forward price difference can be
expressed as s s
F (t, s) - f (t, s) = p vt f S C ~ { e x p [ £ l o g R ( u ) d u ] £ [ S (u)/B(u) ]
{cov[S(u),B(u)] - var[B(u)]}d (u)}}, [6]
where R(u) = 1 + spot interest rate for all u, and
S(u) = spot price of the underlying asset.
Eqs. [5] and [6] suggest that futures and forward prices
differ structurally if future interest rates are stochastic.
They contain implications that are testable in the T-bill
futures market.
Outline
The organization of the remainder of this study is as
follows. Chapter II provides an overview of the T-bill
futures market. Chapter III reviews literature on T-bill
futures market efficiency based on arbitrage arguments and
explains the structural difference between futures and for-
ward prices. Because variance tests of market efficiency
are relatively new, the chapter also provides a review of a
few studies that use the technique. Chapter IV develops
statements of hypotheses. Subsequent sections provide the
methodology to test the hypotheses. Chapter V presents the
results. Chapter VI provides a summary and discusses the
importance of the findings.
CHAPTER BIBLIOGRAPHY
Cox, John, Jonathan Ingersoll and Stephen Ross, "The Relation Between Forward Prices and Future Prices," Journal of Financial Economics, IX (December, 1981) , 321-346.
Fama, Eugene, "Efficient Capital Markets: A Review of Theory and Empirical Work," Journal of Finance, XXV (May, 1970), 383-415.
CHAPTER II
THE TREASURY BILL FUTURES MARKET
Emergence of the Treasury Bill Futures Market
During the early 1970s, interest rates, which had been
fairly stable for years, became highly erratic. The sharp
interest rate swings created a need for price protection.
Some mechanism was necessary to hedge against interest rate
fluctuations. The success of commodity futures prompted
development of financial instrument futures. Such futures
were first applied to the Government National Mortgage
Association securities (GMNAs). A few months later
(January 6, 1976), the International Monetary Market (IMM)
started trading United States T-bill futures. The three-
month T-bill futures contract serves the function of trans-
ferring short-term interest rate risk. It won immediate
acceptance and its growth has been explosive. One reason
for this acceptance is that the yield movements of its
underlying instrument closely correlate with those of other
short-term financial instruments. Changes in the bill
yields reflect interest rate movements free of credit
constraints that affect some other money market instruments.
Consequently, T-bill futures provide hedges against yield
10
11
volatility in T-bills and all other short—term instruments
(e.g., Commercial Paper, CDs, and Bankers Acceptances).
Subsequent introduction of futures trading in other money
market instruments have met very limited success; the T-bill
futures contract seems to have taken care of much of the need
to transfer short-term interest rate risk.
Characteristics of the Treasury Bill Futures Market
This section discusses the specifics of the T-bill
futures contract and the safeguards provided by the organ-
ized exchange on which the futures contract is traded—the
International Monetary Market.
Contract Specifics
A T-bill futures contract is a standardized forward
contract, traded on an organized exchange, to sell or pur-
chase a fixed quantity of T-bills at a set price in the
future. Each T-bill futures contract calls for delivery of
T-bills maturing thirteen weeks from the delivery day with a
par value of one million dollars at bill maturity. The
delivery months are March, June, September and December.
The life of a futures contract can extend up to two years
into the future. At any one time, eight T-bill futures
contracts will be outstanding—eight contract delivery months
at quarterly intervals for eight quarters into the future.
At the delivery date, the seller can deliver any United
12
States T-bill with thirteen weeks to maturity. However,
only a small percentage of contracts traded end up in
delivery. Any trader with a long position who wishes to
cancel it, can do so by taking a short position of the same
amount. Most traders terminate their contracts before
expiration by taking an offsetting position.
Pricing of T-bill futures contracts is based on the IMM
quotation. Because the underlying instrument is quoted on a
yield basis, bids are higher than offers due to the inverse
relationship between yield and price. This quoting method
is contrary to normal market practice where bids are normally
lower than offers. To reduce confusion, the IMM established
a different method of quoting T-bill futures prices. The
IMM quotation is the difference between the T-bill yield
and 100. The IMM method conforms to the traditional quoting
method where bid prices are lower than offer prices. The
quotation, based on an annualized rate, is not the real
price of the thirteen-week spot T-bill. It serves only to
facilitate understanding by traders.
The exchange requires an initial margin deposit of
$1,500 per T-bill futures contract. The limit for daily
price changes is normally sixty basis points above or below
the final settlement price on the previous day. The IMM
also requires a minimum daily price change for T-bill
futures contract of one basis point which changes the value
13
of one futures contract by twenty-five dollars. This study
provides more detailed discussion on minimum margins and
change limits in the next subsection.
Safeguards
To insure that futures markets function smoothly and
free from abuses and disorder, the IMM established safe-
guards .
Clearing House.—The Clearing House is part of the
organized exchange and it serves as the opposite party in
all transactions. After the buyer and seller have initiated
a T-bill futures contract on the exchange floor, the Clear-
ing House acts as a go-between. As a result, no obligatory
relationship exists between the two parties. Each party is
subsequently free to act independently of the other. The
creditworthiness of the opposite party to the transaction is
of no concern because the Clearing House guarantees each
contract. In return, members of the exchange contribute to
a fund which backs the Clearing House.
Settlement of all transactions takes place daily at the
end of the trading day. Before trading can resume the next
business day, each account is "marked to market." The
daily settlement process involves adjustment of accounts to
reflect profits or losses from futures price changes.
Accounts are credited with profits and debited with losses;
14
the net balance of the Clearing House ledger is zero before
trading starts the next day.
Daily price limits.—To limit market participants'
exposure on any trading day, the exchange limits maximum
price changes from the settlement price of the previous day.
Price ranges are determined based on experience in the under-
lying cash market. If the price change in any trading day
is more than sixty basis points above or below the preceding
day's settlement price, the exchange restricts trading; no
T-bill futures trades would be cleared if the contract's
price changes by more than $1,500 (in either direction) from
the previous day's level. If trading restriction applies
for two successive days, widening of the price range is
necessary to accommodate the increased price volatility.
The normal range (sixty basis points) is reinstated if
trading is within the expanded limits. On the last trading
day, the price limit safeguard does not apply.
In addition to limiting participants' daily exposure,
the limits provide the market time to evaluate the causes
of unusual conditions. Manipulative practices may exert
artificial price pressure. The exchange has the authority
to take immediate corrective action in such situations.
Margin requirements.—Margin requirements play a key
role in the clearing system. They insure that market par-
ticipants can withstand adverse price movements. The
15
initial margin ($1,500) is not a downpayment; it is security
to cover losses from price changes that move against the
participants positions. Traders can post the initial margin
either in cash, selected securities, or bank letters of
credit. Of the initial margin, $1,200 is the maintenance
margin. As mentioned, margin accounts are debited (losses)
and credited (profits) when accounts are marked to market
daily. If losses reduce an account below the $1,200 mainte-
nance margin, the accountholder has to bring it back to the
$1,500 initial level (cash only) before trading starts the
next day. If the margin falls below $1,500 but not below
$1,200, the participant does not have to deposit the differ-
ence. Traders can withdraw profits and reinvest the money
elsewhere. The price for determining profits and losses in
margin accounts is the daily settlement price, which is the
average price in the last thirty seconds of trading.
Advantages of Forward Trading in Futures Markets
The T-bill futures market provides a way of hedging
risks. It also provides an opportunity for speculators to
try to make a profit. However, the cash market can also
provide the same functions through a series of transactions.
If so, why is there a need for a futures market? This sec-
tion examines some forward trading techniques and their
drawbacks (1). The discussion should make clear why T-bill
futures are useful.
16
Short selling is one way of forward trading. A short
sale is the sale of borrowed securities (in hope of a price
decline). To hedge security holdings in a short sale,
profits from the sale can offset the drop in value of
security holdings in the case of a price decline. The
ability to hedge security holdings is important to T-bill
dealers. Bill dealers can continue to make a market even
in risky periods by shorting bills. Any dealer constantly
absent from risky markets may create resentment and lose
customers. In the highly competitive bill market, dealers
depend on repeat business.
There are two ways to borrow securities in a short
sale. One is on a demand basis and the other is through a
reverse repurchase agreement for a designated time period.
Borrowing securities on demand subjects the borrower to the
risk of having to close out the position sooner than
expected. The use of a reverse repo—a repurchase agreement
initiated by the lender of funds—can lock in the desired
time period. In this case, the short seller is also subject
to the risks of the repo market which include the credit
risk of the repo seller and the risk of changes in the
T-bill collateral value. If the seller defaults, the buyer
may lose because the T-bill has fallen.
Cash market traders can effect forward trading through
long forward commitments. A long forward commitment is a
17
contract to purchase T-bill securities at a specified date
in the future at an agreed-upon price; it enables a trader
to profit from price increases without holding the securi-
ties. The long forward contract is not standardized and
involves credit risks with regard to the other side of the
transaction. Since such contracts are non-standard, it may
be difficult to locate another party with exactly the oppo-
site need.
Traders can also trade T-bills forward on a when-issued
basis. A when-issued commitment is a promise to purchase or
sell T-bills on delivery day during the period between the
auction of the bills and the settlement day. There is a
time lapse between the acceptance of competitive bids and
the settlement day when buyers make payments to the Federal
Reserve. On a when-issued trade, no cash changes hands
until the completion of the transaction. The major dis-
advantage of trading on a when-issued basis is that it tends
to be short term.
There are several substantial drawbacks to forward
trading in the cash market (e.g., creditworthiness of the
other party, short duration problem, and unequal access of
less active participants to forward trading transactions
or techniques). The T-bill futures market provides the
same function without the accompanying limitations. IMM
safeguards largely eliminate the creditworthiness problems.
18
The most distant T-bill futures contract is two years,
solving the short duration problem. Trading in the futures
market is by open outcry auction. It puts less active,
smaller participants on equal footing with major T-bill
dealers.
CHAPTER III
REVIEW OF LITERATURE
Empirical studies of the T-bill markets reveal a dif-
ference between futures and forward prices. Such studies
generally attribute the difference either to transaction
costs, taxes, the daily settlement process in futures trad-
ing, or a combination of these factors. The unaccounted—for
price difference implies market inefficiency (arbitrage
opportunities). Another approach to evaluate market effi-
ciency is testing for excessive variance in spot and futures
prices. This chapter briefly reviews literature on (1)
futures-forward price differences attributable to transac-
tion costs, (2) futures-forward price differences attribut-
able to the daily settlement process and different tax
treatments, and (3) variance tests of market efficiency.
Futures-Forward Price Difference and Transaction Costs
There have been several studies of the efficiency of
T-bill futures market. They examine the bill futures market
efficiency by comparing futures with the corresponding for-
ward rates implied in the term structure of T-bill interest
rates. Without an explicit T-bill forward market, the for-
ward rate is computed from the appropriate long and short
20
21
interest rates. The discrepancy between futures and forward
rates suggests arbitrage opportunities. Persistent arbi-
trage opportunities imply blatant inefficiency--that the
market is inefficient in the weak form as suggested by Kolb
(10). This section reviews market efficiency tests using
arbitrage arguments.
Poole (15) does one of the earliest T-bill futures
market studies. Poole hypothesizes that the inequality
between futures and forward rates is attributable to higher
transaction costs in the spot market. The implied forward
rate should be higher than the futures rate. Poole investi-
gates the issue by defining the upper and lower critical
points for profitable arbitrage, taking into account com-
mission costs and margin requirements. Profitable arbitrage
opportunities should not exist if futures rates were between
the critical points. Using daily data from January 6, 1976
through June 23, 1977, Poole finds very few profitable arbi-
trage opportunities. Poole's findings support his hypothe-
sis futures rates should be lower than the implied forward
rates. Poole studies only the nearest-to-maturity futures
contract, but he extrapolates his findings to other, more
distant maturity futures contracts. The implication of
Poole's findings is that the T-bill futures market is
efficient in the weak form.
Land and Rasche (11) also compare T-bill futures
yields with corresponding forward yields implied by the term
22
structure. Using Poole's formulas to calculate upper and
lower arbitrage points, they extend the analysis to futures
contracts with longer maturities. Futures maturities are as
long as twenty-four months. Lang and Rasche examine all
eight T-bill futures contracts. T-bills comparable in
length of maturity to the four longest futures contracts do
not exist. Lang and Rasche use yields on treasury coupon
securities to calculate forward rates. Sample data are
collected from March, 1976 through March, 1978. The authors
divide the overall period into three different sub-sample
periods. They find substantial differences between futures
and forward rates for contracts with longer maturities.
Futures rates are higher than forward rates and the rate
differential increases with time to maturity in their sample.
Their findings do not support Poole's hypothesis that futures
rates should be below forward rates.
Lang and Rasche conclude that the substantial differ-
ence between the two rates is not attributable to the ini-
tial low volume in the relatively new bill futures market.
The rate differential has not narrowed consistently over
time. The market has not become more efficient over time.
There appear to be frequent arbitrage opportunities with
longer maturity contracts. Such findings suggest that the
T-bill futures market is inefficient.
Puglisi (16) compares rates of return on two alterna-
tive strategies—futures and outstanding T-bills in the cash
23
market. The bills-only strategy involves purchase of a
short-term T-bill while the bills-futures strategy involves
purchase of a longer term bill while shorting a T-bill
futures contract. The latter strategy involves making
delivery on the futures contract (when it matures) with the
longer term bill. The hypothesis is that if the market is
efficient, the bills-only strategy and the bills-futures
strategy should earn the same return. Assuming a round-trip
commission of sixty dollars per futures contract, the author
finds a significant difference in returns between the two
strategies. Arbitrage profits can be earned—suggesting
market inefficiency. Puglisi's results indicate that market
inefficiency has a tendency to decrease over time, contrary
to the Lang-Rasche's findings.
Capozza and Cornell (2) investigate the efficiency of
the three futures contracts nearest to maturity. Comparing
futures rates with corresponding implied forward rates, test
results indicate significant deviations between rates in the
two markets. Deviations for the near contract are generally
small. The deviations tend to increase with length to
maturity. For the contracts with twenty-six and thirty-nine
weeks to maturity, the differential between futures and for-
ward rates is greater and there is no tendency for it to
decline over time. Moreover, the results show that futures
rates consistently exceed forward rates for those two con-
tracts .
24
Capozza and Cornell argue that arbitrage opportunities
do not necessarily imply that profitable arbitraging is pos-
sible. Arbitrage costs (costs of initiating and closing a
futures position, purchasing and selling spot T-bills, and
short selling in spot bills) could offset any gain. The
cost of short selling, the most significant of the three
costs, is a function of time. Consequently, the differen-
tial between forward and futures rates can increase with
maturity without attracting arbitrageurs. In economic
sense, the T-bill futures market may be efficient even if
arbitrage opportunities do exist.
Rendleman and Carabini (17), henceforth RC, determine
a range within which profitable arbitrage opportunities
would be absent. The bounds of the range are defined as
L = 100P®/P^ - .006, and [7]
U = 100P^/P® + .006, [8]
where L = lower bound,
U = upper bound,
P = spot price per $100 face value for T-bills ill
maturing at time m,
Pn = spot price per $100 face value for T-bills
maturing at time n, and
A and B = asked and bid prices, respectively.
RC analyze the price performance of the first three
futures contracts. A pure arbitrage opportunity refers to
25
a. trader making an arbitrage profit without taking an
initial position. With no pre-existing portfolio of spot
bills, arbitrageurs must short spot T-bills—incurring
heavy borrowing cost.
Taking into account transaction costs, the results
indicate that no profitable pure arbitrage opportunities are
available during the sample period. However, their results
also indicate that profitable quasi-arbitrage opportunities
are frequently available. A quasi-arbitrage opportunity
exists for an arbitrageur who has a pre-existing portfolio
of spot T-bills. RC conclude that the market might not be
efficient. The costs of training traders and policymakers
in financial institutions and of monitoring the futures
market, as well as the reluctance of financial institutions
to change the maturity structure of their portfolios, might
be enough to offset the quasi-arbitrage opportunities.
Futures-Forward Price Difference,
Daily Settlement and Taxes
There are several studies that offer other possible
explanations for the futures-forward differential. Morgan
(14) contends that the price difference results from the
daily marking-to-market mechanism required on futures
contracts. Futures prices could not equal forward prices,
even in an efficient market. The daily settlement process
requires the realization of a profit or loss with the
26
adjustment of the futures price change in the investor's
account at the end of each trading day. In case of a loss,
interest expense is incurred until the delivery date. With
a profit, withdrawing it creates reinvestment interest
income. Morgan's equation in defining the relation between
forward and futures prices is
P D - pd 4. v_1^A-nD-ts D-t for ~ fut ^_1 fut D-t' W
where P°Qr = forward price at D days prior to delivery,
Pf u t = futures price at D days prior to delivery,
= backward difference operator, and
D-t n ^ „
rD-t ~ D _ t day interest rate observed at D-t days
prior to delivery.
This equation shows that forward and futures prices
will not be equal as long as a probability exists that
futures price will change through time. The expected course
of interest rates over the time until the contract expires
is extremely important to futures pricing and it contributes
to the futures-forward differential. The differential can-
not be arbitraged away. Morgan provides no empirical sup-
port for his theory.
Cox, Ingersoll, and Ross (5) derive a model in which
forward and futures prices are unequal when interest rates
are stochastic. Forward and futures markets are fundamen-
tally different. Assuming perfect markets, no taxes and
transaction costs, and the existence of an implicit forward
27
market, CXR formulate a theory that highlights the cause of
the difference—marking-to-market mechanism. In a forward
contract, no payments are made until maturity; the payoff on
a forward contract equals the difference between the forward
price and the maturity spot price. The payoff on a futures
contract equals the difference between the initial futures
price and the maturity spot price. However, the payoff on a
futures contract is received in installments throughout the
period of the contract due to the daily settlement process.
With a forward contract, the total payoff is received at
maturity. The forward contract is comparable to investing
in a long-term security that matures at the same time as the
forward contract. However, the futures contract is compara-
ble to reinvesting continuously in one-day securities until
the futures contract matures. Forward and futures contracts
are dissimilar if interest rates are stochastic because the
futures price will then be a function of the unknown one—day
interest rates over the period of the futures contract. The
futures-forward price differential should not be interpreted
as evidence of market inefficiency, but rather as a result
of structural difference between futures and forward con-
tracts. This price differential cannot be arbitraged away.
Jarrow and Oldfield (9), henceforward JO, independently
reach the same conclusion as CIR—the forward price does not
equal the futures price unless default-free interest rates
28
are nonstochastic. Forward and futures contracts are
deceptively similar, but the difference arises due to the
daily marking-to-market mechanism in the futures market.
JO use a series of arbitrage positions to show that the
forward price is equal to the futures price only if interest
rates are deterministic.
JO also argue that the relationship between forward and
futures contract values depends upon the default-free rate.
JO define the relationship between forward and futures con-
tract values as follows:
f (p(s,t*) ;k(t) ft*) = F(P(s,t*) ?K(t) ,t*) .B(s,t*) , [10]
where f = forward contract value at time s,
F = futures contract value at time s,
t* = delivery date,
k(t) = exercise price of forward contract,
p(s,t*) = forward price at time s,
K(t) = exercise price of futures contract,
P(s,t*) = futures price at time s, and
B(s,t*) = spot price at time s of $1 received with
certainty at delivery date.
If the default-free rate is deterministic, the equation
above describes the relationship between forward and futures
values. As the equation shows, the forward contract value
is a proportion of the futures contract value at time s.
The discrepancy between f and F is due only to the different
29
cash flow patterns. Therefore, a futures contract is not
equal to a forward contract rewritten at the end of each day
even though the default-free rate is deterministic and the
forward and futures prices are equal. If the default-free
rate is stochastic, there is no relationship between f and
F.
Both CIR and JO show that futures and forward prices
are not equal if interest rates are stochastic. Neither
provides empirical evidence. French (7) empirically
examines two of the CIR propositions. According to CIR,
the futures-forward price difference should be positive if
the covariance between the daily percentage changes in the
futures price and the bond price is negative, and vice versa.
Further, the futures-forward price difference should be
negative if the bond price variance is smaller than its
covariance with the spot price, and vice versa. French uses
futures and forward prices for silver and copper to test
these two CIR propositions.
French encounters several serious complications in his
empirical work. Prices are collected from 1968 through
1980, which includes the years of 1979 and 1980 when the
economy experienced astronomical interest and inflation
rates. Another problem is matching futures and forward
prices—the futures prices reflect trading on the Commodity
Exchange and the Chicago Board of Trade in the United States
30
and the forward prices reflect trading on the London Metal
Exchange in Great Britain. Trading time differences might
introduce serious measurement errors; prices in Great
Britain are observed several hours earlier than those in the
United States. Other problems are changes in exchange rates,
transportation costs and international trade restrictions.
French finds some support for the first proposition;
support for the second proposition is more substantial.
Generally, the evidence is consistent with the CIR pre-
dictions, suggesting that the observed difference between
the forward and futures prices may not be evidence of market
inefficiency. This difference may be caused by structural
differences between futures and forward contracts. It is
unclear whether such conclusions apply to the T-bill futures
market.
Another explanation for the futures-forward differen-
tial is tax treatment. Several studies investigate whether
the difference between the T-bill futures price and the
implicit forward price might be attributable to taxes.
Cornell (3) employs a methodology that does not require
computation of implicit forward prices. The methodology is
similar to the one used to study the effect of taxes on
stock prices—examining returns on ex-dividend date when no
dividend is available to be taxed as ordinary income. In
November, 1978, the Internal Revenue Service ruled that all
31
futures contracts would be considered capital assets. The
holding period for long-term gains is six months for long
positions while all gains and losses on short positions are
considered short-term, regardless of the length of the
holding period. This asymmetric tax treatment should
encourage investors to take long positions with over six
months to maturity. If gains occur on the last trading day,
immediately liquidating futures contracts would lock in long-
term gains. When losses exist, taking delivery of the bills
at maturity and immediately reselling the bills in the cash
market establishes ordinary losses.
The tax ruling makes taking a long position in bill
futures with a maturity of more than six months attractive.
If taxes affect the futures-forward differential, futures
prices should drop when the maturity of bill futures con-
tracts changes from six months and one day to exactly six
months. Using a sample period of September, 1976 to March,
1980, the empirical study finds no tendency for futures
prices to drop when the critical maturity is reached.
Arak (1) also examines the effect of differential tax
treatment of bill futures and implicit forward contracts.
The sample period is January, 1976 to March, 1982. By
comparing futures and forward rates, Arak finds that the
November, 1979 IRS ruling has a significant impact on
futures prices. After the ruling, prices of more distant
32
maturity contracts rose significantly above forward, prices.
Even after the passage of the Economic Recovery Tax Act of
1981 (ERTA), the spread between the futures and forward
prices persisted. ERTA requires that for all positions
established after June 23, 1981, any net gain is taxed as if
60 percent is long term and 40 percent is short term, regard-
less of the holding period. ERTA also treats T-bills as
capital assets. As a result, taking delivery in order to
convert capital losses into ordinary losses is no longer
possible. However, there is insufficient data to draw firm
conclusions on the impact of ERTA.
Most empirical studies of the T-bill futures market
conclude that the difference between futures and implied
forward prices is attributable to either transaction costs,
marking-to-market procedures, or taxes. Cornell and Reinga-
num (4) examine all three factors by using data from the
foreign exchange market. Using a sample period of June,
1974 to June, 1979, the authors collect data on six major
currencies.
The empirical results indicate that the discrepancy
between forward and futures prices observed in the foreign
exchange market is not similar to the discrepancy found in
the T-bill market. The foreign exchange data show the mean
difference to be significantly different from zero from both
statistical and arbitrage points of view. Since the CIR
33
effect is negligible in both the foreign exchange and T-bill
markets, the only remaining explanation for the T-bill
results are differential tax treatment and transaction costs
unique to the T-bill futures market. These transaction
costs include the cost of short selling (due to the absence
°f a T-bill forward market), Cornell and Reinganum do not
attrioute the difference between the T—bill futures and
implicit forward prices to market inefficiency; they suggest
tax treatment and transaction costs.
Variance Tests of Market Efficiency
A traditional test of market efficiency has been to run
a regression to determine whether innovations are forecast^
able. If a market is efficient, only the arrival of unantici-
pated information results in price changes. Arrival of new
information is random so that price changes are not fore-
castable.
Recently, a few studies have used a different approach
to evaluate market efficiency. The studies use volatility
measures to determine whether prices show excessive varia-
tion to be accounted for in terms of the random arrival of
new information. The variance tests are a relatively new
efficient markets methodology. A chronological review of
some studies using this approach seems appropriate.
Shiller (18) was the first to use variance tests to
test bond market efficiency. The rational expectations
34
theory implies that a long-term interest rate is an average
of expected short-term rates. Therefore, long-term rate
series should be smoother than the short-term rate series.
The variance of the long-rate series should be limited by
the variance of the short rate series. It has been sug-
gested that long-term interest rates in rational expecta-
tions models are too volatile to be explained in terms of
the averaging inherent in the models. If long—term interest
rates are too volatile, it implies that the market is inef-
ficient. The conditional average (long rate), representing
rational expectations of future short-term rates, should
change only with arrival of significant new information.
However, if the long-term rate movements are "too big" rela-
tive to actual subsequent events, the movements cannot be
realistically attributed to new information. If this phe-
nomenon occurs too often, it implies a kind of forecastabil—
ity for long rates. In such case, public expectations are
not rational as expounded by expectations models.
Shiller uses the bond market, generally regarded as
efficient, to show that long-term rates are too volatile to
be consistent with rational expectations theory. The vola-
tility of long—term rates, as measured by the variance of
short-term holding yields on long-term bonds, should not
exceed bounds imposed by expectations models. Shiller finds
that the holding-yield variance generally exceeds its
35
theoretical oounds. This excess volatility implies that
either* (1) the expectations models of the term structure
are too restrictive, or (2) economic agents do not form
^^tional expectations and the bond market is inefficient.
Shiller (20) also examines whether stock prices move
more tnan justified by subsequent changes in dividends.
According to Shiller, the simple efficient markets model
that relates stock price to rationally—expected future divi-
dends is as follows:
CO
P t = k=o t < 1 / 1 + r > k + l E t < d t + k' }' 1111
where P = stock price at time t,
Dt+k = d i v i d e n d a t time t+k,
= mathematical expectations operator conditional
on information at time t, and
r = constant discount rate.
If stock prices equal the present value of expected
future dividends, any movement in stock prices is attribut-
able to random arrival of new information about future
dividends. Shiller finds that the volatility of stock price
is five to thirteen times too high to be attributed to new
information about future dividends. The upper bound, repre-
sented by the measure of the observed variability in divi-
dends, has been so dramatically violated by stock price
volatility that it would be unfair to attribute the failure
of the efficient market model to problems such as data
36
errors. Shiller argues that if the stock market is
efficient, the sample standard deviation of the movements
of dividends must understate the real uncertainty about
future dividends. The stock market might have rightfully
feared larger movements of future dividends than are actu-
ally realized.
Shiller (19) compares regression and variance tests
for evaluating market efficiency. He suggests several
advantages that volatility analysis has over regression
analysis. One of them is that a volatility test is more
robust and simple. Shiller further justifies the use of
volatility measures in assessing market efficiency because
they are insensitive to the problem of data misalignment.
In assessing the efficiency of the stock market, the price
movements during the sample period may reflect new data
about dividends that materialize after the sample period.
On the other hand, price movements reflecting information
that occurs during the sample period may take place before
the sample period. Regression tests are impossible when the
data are totally misaligned. Unaligned observations do not
affect variance tests. Shiller implies that variance tests
are more powerful than regression tests.
LeRoy and Porter (13) independently derive market
efficiency tests based on implied variance bounds by using
a present-value relation model. The model is as follows:
37
n Y = Z B3x®(j), [12]
j=0 ^
where Y = the stock price series,
0
xt (j)
= E^xt+j//lt^ wfiere It i
s the information on xt
and zt up to and including time t, = the earning series,
z - = a nY variable other than past earnings that are
used to predict earnings, and
B = discount factor and less than one.
Using the above model, the authors come out with three
theorems about the variance of the dependent variable process
relating to the variance of the independent variable process.
Theorem 1 states that the coefficient of dispersion of
Yfc is less than the standard deviation-mean ratio found in
• The theorem is theoretically sound because the present-
value model defines stock price as a weighted average of
expected earnings. With the average smoother and less vola-
tile than its various components, and the tendency of
expected earnings to move toward a mean in the increasingly
distant future, the coefficient of dispersion of stock price
should be relatively less.
Theorem 2 states that
CD of Y* > CD of Yfc >_ CD of Y , [13]
where CD = coefficient of dispersion, *
Yt = stock price series if future earning changes
are known with certainty.
38
Yfc = stock price series defined in Eq. [12], and
~ stock price series if future earning changes
are unknown completely.
By definition, equals (Y_ + U .) ; U_ is the forecast error
depending only on earning changes occurring after time t.
If the market is efficient, the forecast error cannot be
correlated with Yt—any correlation implies a biased fore-
cast of Yt> Assuming market efficiency, var(Y ) = var(Y ) + t t
var(Ut) if the covariance between Y and U is zero. It
follows that var(Yt) > var(Yt); the variance of Y* estab-
lishes the upper bound for the variance of Y . Likewise, ^ A
var(Yt) = var(Yt) + var (Ut). Assuming Yfc contains at least
the information on the past history of earnings, var(Y ) > A t —
var(Y^) ' the variance of Y_ establishes the lower bound on
the variance of Y .
Theorems 1 and 2 are similar to the restrictions that
Shiller obtains on the variance of long rates. Theorem 3
has already been stated in discussing Theorem 2. To restate
the first part of Theorem 3, the variance of Y equals the
sum of the variance of Y and U . L. L
The three theorems can be used as tests of market
efficiency because the present-value relation between stock
price and earnings implies market efficiency. Using a data
set based on the Standard and Poor's Composite Index of
stock prices and related earnings, the authors test all
39
three theorems. They find a clear rejection of market
efficiency during the sample period, 1955 to 1973. Stock
prices seem to be more volatile than is consistent with
market efficiency. This is contrary to the large body of
market efficiency literature using regression to determine
the presence or absence of autocorrelations in rates of
return. Since the autocorrelation tests and volatility tests
are derived from the same basic model, their different out-
comes regarding market efficiency puzzle the authors. One
explanation could be that the volatility or dispersion tests
have more power than the regression or autocorrelation tests.
Regression and variance tests of market efficiency do
not always produce conflicting results. Huang (8) studies
the efficiency of the foreign market. He rejects the null
hypothesis of market efficiency with both the regression and
variance tests. The model examined is the monetary approach
to exchange rate determination. Market efficiency tests are
necessarily joint tests of market efficiency and model
validity. Rejecting the null hypothesis does not necessarily
mean that the foreign exchange market is inefficient, but
that the model is misspecified.
Studies in asset-price volatility suggest that price
variance is too great to be consistent with conventional
security valuation models. LeRoy and LaCivita (12) ques-
tion these variance test conclusions. They argue that the
40
conventional models are based on risk-neutrality or near
risk-neutrality. Risk-aversion leads to greater asset price
volatility than risk neutrality. It is then possible that
if risk-aversion is taken into account, more volatility is
permitted which might change interpretation of the results
on asset-price dispersion.
LeRoy and LaCivita point out that efficient market
tests are joint tests of three assumptions: stationarity,
rational expectations, and present-value relation. A
restricted (two-moment) stationarity assumption requires
that
var(Pt) =var(P t + 1), [14]
where P = price or yield, and
var (.) = variance operator.
The rational expectations assumption requires the opti-
mal forecast of future spot prices to be made, based on
available information. The present-value relation assump-
tion is the element of the joint hypothesis that the authors
emphasize in their study. The assumption is equivalent to
the martingale assumption (that the future spot price is
equal to the current spot price) in the stock market. They
maintain that the assumption is valid only under risk-
neutrality . LeRoy and LaCivita conclude that there is a
positive connection between risk-aversion and price volatil-
ity. It still remains an open empirical question as to what
41
degree of risk-aversion is adequate to account for the
observed volatility in asset prices.
Flavin (6) also reassesses the empirical evidence
regarding excess volatility in the financial markets. The
seeming violation of market efficiency using variance mea-
sures may simply reflect small sample bias. Variance tests
tend to be biased toward rejecting the null hypothesis of
market efficiency. In variance analysis, variances are
expressed in deviations from the sample mean rather than the
population mean. Obtaining deviations from the sample mean
creates a downward bias in the sample variance, due to
asymptotic sample properties. Thus, implied upper variance
bounds should be higher if the population means were known a
priori.
Summary
Previous studies offer various explanations for the
price discrepancy in the T-bill spot and futures markets.
In general, pure arbitrage opportunities are rare. However,
observation of quasi-arbitrage opportunities are frequent—
possibly due to the absence of the short selling portion of
transaction costs. The futures market is inefficient in the
quasi-arbitrage sense.
A competing explanation for the futures-forward price
differences is the daily marking-to-market feature in the
futures market. Morgan argues that such a feature would
42
create futures-forward price differences even in an
efficient T-bill futures market. Cox-Ingersoll-Ross and
Jarrow-Oldfield both provide theoretical justification for
the futures-forward price differences: they could be attrib^
utable to stochastic interest rates. French obtains empiri-
cal support for two of CIR's propositions in a study of
copper and silver, but some serious data problems undermine
the effectiveness of the study.
A third explanation for the T-bill futures-forward price
differences is the differential tax treatment. Cornell finds
no evidence that taxes have contributed to the futures-
forward differential. Arak finds evidence to the contrary.
In an effort to determine which of the three competing
explanations for the bill futures-forward price differential
is most promising, Cornell and Reinganum examine all three.
They suggest that only the differential tax treatment and
transaction costs explanations seem reasonable. The effect
of the daily marking should be negligible.
The differences between futures and implicit forward
prices suggest arbitrage opportunities. Unsatisfactory
explanations for the differences infer market inefficiency.
The conflicting findings of previous studies leave the issue
of T-bill futures market efficiency unresolved,
Variance-bound tests suggest excess volatility in the
financial markets. Shiller points out that long-term rates
43
and stock prices are too volatile to be consistent with
their corresponding models. LeRoy and Porter conclude that
excess stock price volatility contradicts the market effi-
ciency hypothesis. Huang finds the foreign exchange market
inefficient. These findings suggest that either the models
at hand are not valid or the markets studied are inefficient.
However, both LeRoy and LaCivita, and Flavin caution that
apparent evidence of market inefficiency may be attributable
to the unrealistic assumption of risk-neutrality (which is
the basis for conventional security valuation models) and
small sample bias.
This chapter reviewed literature on theoretical
futures-forward price differences and efficiency tests based
on arbitrage arguments. The issue of T-bill futures market
efficiency needs further examination using different method-
ological approaches. The next chapter describes the alterna-
tive methodologies. The methodology chapter also includes
operational statements on the empirical testing of the T-
bill futures-forward difference.
CHAPTER BIBLIOGRAPHY
1. Arak, Marcelle, "The Effect of the Tax Treatment of Treasury Bill Futures on Their Rates," Journal of Futures Markets, III (Spring, 1983), 65-73.
2. Capozza, Dennis and Bradford Cornell, "Treasury Bill Pricing in the Spot and Futures Markets," Review of Economics and Statistics, LXI (November^ 1979), 513-520.
3. Cornell, Bradford, "A Note on Taxes and the Pricing of Treasury Bill Futures Contracts," Journal of Finance, XXXVI (June, 1981), 1169-1176.
4. Cornell, Bradford and Marc Reinganum, "Forward and Futures Prices: Evidence from the Foreign Exchange Markets," Journal of Finance, XXXVI (December, 1981), 1035-1045.
5. Cox, John, Jonathan Ingersoll and Stephen Ross, "The Relation Between Forward Prices and Futures Prices," Journal of Financial Economics, IX (December, 1981), 321-346.
6. Flavin, Marjorie, "Excess Volatility in the Financial Markets: A Reassessment of the Empirical Evidence," Journal of Political Economy, XCI (December, 1983), 929-956.
7. French, Kenneth, "A Comparison of Futures and Forward Prices," Journal of Financial Economics, XII (November, 1983), 311-342.
8. Huang, Roger, "The Monetary Approach to Exchange Rate in an Efficient Foreign Exchange Market: Tests Based on Volatility," Journal of Finance, XXXVI (March, 1981), 31-41.
9. Jarrow, Robert and George Oldfield, "Forward Contract and Futures Contracts," Journal of Financial Economics, IX (December, 1981), 373-382.
10. Kolb, Robert, Interest Rate Futures: A Comprehensive Introduction, Richmond, Virginia,-Robert F. Dame, Inc., 1982.
44
45
11. Lang, Richard and Robert Rasche, "A Comparison of Yields on Futures Contracts and Implied Forward Rates," Federal Reserve Bank of St. Louis Review, LX (December, 197B), 21-30.
12. LeRoy, Stephen and C. J. LaCivita, "Risk Aversion and the Dispersion of Asset Prices," Journal of Business, LIV (October, 1981), 535-547.
13. LeRoy, Stephen and Richard Porter, "The Present Value Relation: Tests Based on Implied Variance Bounds," Econometrica, XLIX (May, 1981), 555-574.
14. Morgan, George, "Forward and Futures Pricing on Treasury Bills," Journal of Banking and Finance, V (December, 1981), 483-496.
15. Poole, William, "Using T-Bill Futures to Gauge Interest-Rate Expectations," Federal Reserve Bank of San Francisco Economics Review, Spring, 1978, pp. 7-16.
16. Puglisi, Donald, "Is the Futures Market for Treasury Bills Efficient?," Journal of Portfolio Management, IV (Winter, 1978), 64-67.
17. Rendleman, Richard and Christopher Carabini, "The Efficiency of the Treasury Bill Futures Market," Journal of Finance, XXXIX (September, 1979), 895-914.
18. Shiller, Robert, "The Volatility of Long-Term Interest Rates and Expectations Models of the Term Structure," Journal of Political Economy, LXXXVII (December, 1979), 1190-1219.
19. , "The Use of Volatility Measures in Assessing Market Efficiency," Journal of Finance, XXXVI (May, 1981), 291-304.
20. , "Do Stock Prices Move Too Much to be Justified by Subsequent Changes in Dividends?," American Economic Review, LXXI (June, 1981), 421-436.
CHAPTER IV
METHODOLOGY
This chapter begins by developing the hypotheses to be
tested. Subsequent discussion focuses on the procedures and
data used to test the hypotheses. The final section of the
chapter is a summary.
Hypotheses
Futures-Forward Price Difference and Stochastic Interest Rates
Eqs. [5] and [61 show the futures-forward price differ-
ences in terms of the relations between futures and bond
prices, and between spot and bond prices. The futures and
forward prices are equal only if interest rates are non-
stochastic. These models assume no taxes or transaction
costs.
Eq. [5] implies that if the local covariance (localized
to individual contract periods) between the percentage
changes in the futures price and the percentage changes in
the bond price is positive, the integral m the equation
positive. This makes the right side of the equality sign
negative; the futures price should be less than the forward
price. If the covariance is negative, the futures price
46
47
should be higher than the forward price. To test this
hypothesis, the average futures-forward price difference
is compared with the average covariance between the futures
and bond percentage price changes. The null and alternative
hypotheses for a significant mean price difference are
H q: diff(F, f) = 0
H;l: diff(F,f) > 0, and
H 2: diff(F,f) < 0,
where diff(F,f> is the average futures-forward price differ-
ence. The null and alternative hypotheses for a significant
mean local covariance are
HQ: cov(F,B) = 0,
: cov(F',B) > 0, and
H2: COV (F,B) < 0,
where cov(F,B) is the mean covariance between the futures
and bond percentage price changes.
With unknown population standard deviation and the
small sample size of local covariances, the appropriate
statistical test to use is the Student's t. The study
assumes normal distribution for the population values.
The significance level is at 5 percent. ihe definition
the sampling distribution for t is
t = (x - 0)/(s//n) with n-1 degrees of freedom,
where x = variable mean,
s = standard deviation, and
n = number of observations.
48
The decision rule is to reject the null hypothesis if t is
less than the left critical value or more than the right
critical value.
Eg. [6] implies that if the local variance of the bond
percentage price changes is larger than the local covariance
between the spot and bond percentage price changes, the
futures-forward price difference should be positive. If the
variance is smaller than the covariance, the futures-forward
price difference should be negative. Testing this hypothesis
requires comparing the average futures-forward price differ-
ence with the difference between the mean bond price variance
and the mean spot-bond covariance. The null and alternate
hypotheses for a significant mean bond price variance are
HQ : var(B) = 0,
H i var(B) > 0, and
H2*. var(B) < 0.
The hypotheses for a significant mean spot-bond covariance
are also stated likewise. The null hypothesis testing
procedure associated with Eq. [5] also applies for testing
the above null hypotheses.
The empirical result supports the model in Eq. [5] if
diff (F,f) is significantly different from zero and positive,
and cov(F,B) is significantly different from zero and nega-
tive. The same is true if these conditions are reversed.
There is also empirical support for the model m Eq. [6] if
diff (F,f) is significantly different from zero and positive,
49
and [cov(s,B)-var(B)l is significantly different from zero
and negative. The same is true if the signs are reversed.
Variance Tests of Futures Market Efficiency
This subsection begins with the derivation of variance
inequalities. The development of hypotheses follows.
Derivation.—The variance tests of market efficiency
assume weak stationarity in the variables analyzed. Weak
stationarity implies that there are no systematic changes m
the variance. A process is strictly stationary only if all
its moments are constant—same probability distribution fox-
all time. This study assumes only weak stationarity.
Eqs. [2] and [3] in Chapter I provide the basis for the
derivation of two inequality equations that are used as
variance tests of market efficiency. Given Eq. [3], an
efficient market implies that
c°v ( Ft,n
Ut,n> = °' 1 1 5 1
where F = futures price quoted at time t for delivery t f n
at time n, and
U = forecast error given the available information t,n
at time t,
since the forecast error is orthogonal to any element of the
information set. Computing the variance for each variable
in Eq. [3] gives
var(Sn) = var(Ft,n} + var(Ut,n)* [161
50
Since positive semi-definiteness requires var(U^n) > 0, it
follows that
var(S ) > var(F ).
n — n
The futures price, an estimate of the expected spot price on
the maturity date of the contract, should be less volatile
than the realized spot price. Theorem 3 in Leroy-Porter
proposes that var(Yfc) is equal or less than var(Yt), where
Y* represents a time series that contains more information
on the future expected spot price than Yt< Market partici-
pants possess more information at time t+1 than at time t on
the future spot price at time n. It can be expressed as (Sn ' Ft,n' = (Sn " Ft+l,n> + U' 1 1 8 1
where U = additional forecast error due to less information
at time t than at time t+1.
Following the manner in which Eq. [17] was obtained,
var(Sn - F t f n) > var(Sn - F t + 1, n) • [ 1 9 ]
The greater the time distance between the date when the
futures price is quoted and the contract maturity date, the
less information there is to estimate S n and the bigger the
forecast error (as measured by the variance); i.e.,
v a r ( sm ' Ft,m» i v a r ( S n ' Ft,n>' n < m 1201
Following Huang (1), Eq. [2] can be used to derive
another variance inequality. S t is known at time t. If
markets are efficient,
< Ft,n - St> = Et ( Sn ' St>" U U
51
Taking the variances of both sides, Eq. [21] can be general-
ized to obtain the following inequality.
v a r ( Sn'" St ) ~ v a r ( Ft,n ' St) -
var(Sn-St) - var(F
t+i,n ~ St) ' 0 < i < n '-221
With the variability of ( Ft +i ; n ~
st^ a c c o u n ti n <? f°r a
greater portion of the variability of the spot price at time
n,
v a r ( Ft +i,n " St' > v a r ( Ft,n " St»" 1231
Hypotheses for variance tests.—This study uses Eqs.
[20] and [23] to test market efficiency. The null and
alternative hypotheses associated with Eq. [20] are
K0 var(S - F ) > var(S - F ), and
Ha : v a r ( Sm " Ft,m) < v a r ( Sn " "
This study uses the Pitman statistic to test the null hypoth
esis. The test (3) is defined as
R = <si/s2 " X ) /y/ (S2/S2 + l)2 - ^ ( S ^ / S p
with n-2 degrees of freedom,
where S2 = sample estimate of variance,
S2 = sample estimate of smaller variance, and
r = correlation coefficient between two variables.
The statistic is used for a one-tailed test. The probaoility
of committing a Type One error is 5 percent. The null (mar-
ket efficiency) hypothesis is rejected if the computed R is
52
less than the critical R value. The null hypothesis is not
otherwise rejected.
The null and alternative hypotheses associated with Eq.
[23] are
Hq: var<Ft+i(I1 - Sfc) > var(Ft_n - Sfc) , and
Ha: var(Ft+i,n ' St> < var(Ft,n " St>"
The above null hypothesis is tested using the same testing
procedure as described earlier.
Regression Tests of Futures Market Efficiency
Prices in an efficient market reflect all available
information. Only the arrival of unpredicted information
can make traders adjust their forecasts. The futures price,
an estimate of the future spot price and conditional on the
information set at time t, should be unbiased if the market
is efficient. Eq. [2] in Chapter I describes the relation-
ship between the futures price and the expected spot price.
As Eq. [4] has shown, Eq. [2] is an efficient market model
if the forecast error is orthogonal to any element of the
information set. Putting it differently,
VSn " -WV " °' U 4 1
The same is true for futures prices that are quoted at
different dates of the same contract period; they are all
unbiased estimates of the same future spot price.
53
Empirical testing of futures market efficiency can take
the following regression form:
S = a + bF. _ + u . [25] n t, n t
where a = intercept term,
b = slope coefficient, and
u = S - E (S ) when E (S ) = F . t n t n t n
Assuming that ufc are dependent and identically distributed
with zero mean and a2, the null and alternative hypotheses
are
Hq: b = 1, and
H : b f 1. a
This study uses the t statistic to test the null hypothesis
at the 5 percent level of significance. The decision rule
is to reject the null hypothesis if the computed t value is
smaller (greater) than the lower (higher) critical value.
Failure to reject the null hypothesis implies that the
futures market is efficient.
Procedures
The scope of this study is limited to analyzing the
first three T-bill futures contracts. Going beyond the
third contract necessitates the use of Treasury coupon
securities that might introduce measurement problems. The
next subsection presents the procedure for testing the CIR
propositions regarding the difference between futures and
implicit forward contracts. The last chapter subsection
54
deals with the procedure for testing market efficiency in
the three T-bill futures contracts nearest to maturity.
Tests of the Cox-Ingersoll-Ross Models
•phe futures—forward price differences are computed from
price data during the first thirteen-week periods of the
three-, six- and nine-month contracts to avoid overlapping
data. This study defines the futures-forward percentage
price difference as
(Ft " ft ) / ft x 1 0 0' [26]
where = futures price at time t, and
f = implied forward price at time t.
Expressing the price difference in percentage gives the
relative magnitude of the difference. The average futures-
forward price difference is obtained for each of the three
contracts.
Estimating the mean covariance consists of two steps.
First, the local covariance of a given maturity-period con-
tract is estimated over each contract period in the sample.
Second, averaging local covariances across contract periods
estimates the mean covariance. The mean local covariance of
the daily percentage changes in the bond price is obtained
similarly. Subsequent comparisons of the estimated vari-
ables would then determine whether there is support for the
two CIR models.
55
This study further examines the ability of the CIR
theory to explain T-bill futures-forward price differences.
Although the ability of the theory to discriminate among
price differences is not one of the issues in this study,
it is useful to determine whether their theory explains
intra-sample variations. The procedures in this study are
similar to those used by French; he examined the models'
usefulness in describing intra-sample variations in the
copper and silver markets. After the estimation of local
covariances over each three—, six- and nine—month contract
period, the contracts are then assigned to two groups.
Using the first CIR model, the basis of assignment is whether
a contract has a negative or positive covariance estimate.
If changes in covariance and price differences are related,
there should be a positive futures-forward mean price differ-
ence for the group with negative covariance estimates. The
reverse should also be true. A second CIR proposition is
tested in similar fashion. The futures-forward price differ-
ences should be positive in the group where the bond price
variance is greater than the bond-spot price covariance.
The reverse should also be true.
Market Efficiency Tests
This study tests Eq, [20] for n = 13 and 26 weeks, m =
n+13 weeks and t = 1, 4, 7, 10 and 12 weeks. For a given n,
the variance of the absolute price differences between the
56
expected T-bill spot price and the futures price at time t
is computed for each T-bill futures contract period in the
sample. Based on Eg. [20], the mean variances on both sides
of the inequality sign are then compared.
This study tests Eg. [23] for n = 13, 26 and 39 weeks
and t = 0, 3, 6, and 9 weeks. The subscript of i in the
equation represents a period length of four weeks. The pro-
cess of finding the average variance of the absolute price
difference (futures price - bill spot price at time t) is
similar to that used for Eq. [20].
A second approach to test futures market efficiency
involves performing regression tests for n = 13, 26 and 39
weeks and t = 1, 4, 7, 10 and 12 weeks. The expected spot
price at the futures contract maturity date is regressed
against the futures price at time t. In addition to calcu-
lating significance statistics for the regression coeffi-
cients, R values (and F tests) and Durbin-Watson statistics
are reported.
Data
The Wall Street Journal is the source of weekly futures
price data. This study uses the Wednesday closing prices as
reported in the Journal: since the last day of trading is
two business days after the third weekly T—bill auction dur-
ing the contract month for any specific futures contract,
the spot bill auction usually falls on Monday and futures
57
trading for any contract usually terminates on Wednesday.
Moreover, the choice of Wednesday's prices avoids roost
holidays. To maintain compatibility with futures prices,
T-bill spot prices are also computed from Wednesday's clos-
ing bid yields as reported in the Wall Street Journal.
This study uses weekly bid prices because practitioners
assert that bid prices more accurately reflect market prices
The sample period begins March, 1976 and runs through
December, 1984. The overall period is divided into three
subperiods to facilitate more in-depth analysis of the
sample data. Table I shows the length of each subperiod.
T—bill futures trading started on January, 1976, but the
first two months of trading v/ere relatively light. Data
from the first two months of trading are, therefore, omit-
ted. This study also omits data in cases where no spot
T-bills matured at or near the maturity date of the T-bill
futures contract concerned.
Data used in testing the CIR models consist of the
weekly Tuesday-Wednesday price changes, except for those
used in finding the futures-forward price differences. The
bond percentage price change is obtained by measuring the
percentage price change in the spot T-bill that matures at
the same time as the relevant futures contract. The
measurement of percentage changes in the spot T-bill that
matures thirteen weeks after the relevant futures contract
58
maturity date provides the spot percentage price change data
for this study. Wednesday's T-bill bid quotations provide
the basis for calculating implicit forward prices. The
traditional formula to compute implicit forward prices
involves extensive calculations. Roll's formula (2) is used
in this study:
„ fQ1 = 100 - [nRT - (n - 91)R_]/91, [27] n-m 91 L b
where f~, = forward price on a 91-day T-bill to begin n-m 91 r
in n-m periods,
n = number of days to long bill maturity date,
m = number of days to short bill maturity date,
R = spot rate on long bill, and Li
Rc = spot rate on short bill.
Roll's formula may generate different, but inconsequential,
forward prices.
Before generating the results and analyzing them, it is
helpful to examine the general behavior of the sample price
data. Table I presents the comparison of the behavior of
the futures, forward, bond and spot prices. The sample
prices were lowest during the second subperiod (December,
1978 to December, 1981). They were highest during the first
subperiod. The variance statistics indicate that the prices
varied the most during the second subperiod. The smallest
variance value is 3.6 in the nine-month futures contract.
59
W
CQ <
EH
P O H
W
P M
PA
P-I
3 CN
O S3 H
13 Q
& O H
s UJ W PQ < EH <
P
-P tn C A)
> I -P •H U 3 -P ?D 2
rH * fd a (D 0 •H i—i •H -P X* U V1 fd CD •H •H fl4 -P U fd fd >i •P > eQ cn
R-
LO
CO LO cri r-<£>
KD LO
M CO VO
00 O H <T\ O
o ro OO
ro i—I i—I <y\ o
LO
KO o
CM o
* CTI * ro CN RH H
LD CO
R O O H o C^-
00 cn ro o «H cn ko cn
RO CN
KO 00
00 (J\ I H LO V£5
00 rH CN
00
00 O>
R-OO o H
00 cn rH CN
00 I—I
00
^ ^ H cn cn
OO r-
00
ro oo «H cn ro cn
cn ro oo cn r-cn
OO ro
OO ro CN cn LO
CO O LO 00 o
cn o CN oo r-
VP
OO CN
cn cn ro cn
OO CN ^ oo
o in H cn o
OO
CN ro ro
I—1 o cn
<y\ cn xr oo in ro
O kO r-H cn r-
VD CN VO m o\ o ro 00 RH
cn cn CN 00 CN
00 CN
as as ro CS
OO CN oo
A !
O LO I—I cn o
r-CN
00 ro
OO CN KD 00 LO
CN r-• • o ro CN cn o
00 R-cn
OO cn
ro rH
oo <sO ro oo cn
cn in CN CO r-
<X> CH CN
r-V£>
00 CN ^ 00 KD in
cn ro CN 0 0 i—I r-
R-CN
00 ro
CN CN
CO R-
00 CM KD 00 LO
o ro CN cn o
VO CN
I—! O
oo cn 00 CM
r-r- o • • o r- ro
10 CM OO O
00 CN
co m oo ro
cn oo ro oo ro
in I > R^
r-rH
in R-
00 00 LO oo r-
o cn CN cn cn
\D
Q\ L>
O VD VD cn
V£>
r- KD o
o i> r-cn in vo
CM LO
o in R-KD
in ^ o o in cn
OO
r- VO O
cn CM
VO
O
VO R-
o cn vo
U 0 *H U 0 A
X* 3 cn
4J W I—I
-p RD
C 4J rd cn 0) S -P
Q) O C fd
*H U rd >
O •H U CD CU
RQ
Q) • O
-P £ fd fd
D -P *H ui fd cn u
CD fd 2 -p > RT3
FL
CM
T5 0 •H U CD A
XI
U)
U ro
-P fd
C -P fd cn 0) G -P
CD O CJ fd
*H U fd >
13 O •H u CD P4
CD A fd
P 4 CD G S fd
w
A CD
o r-cn in
VJQ
P I
ro r- o o ro cn cn VD
VP
cn o ro in cn RH r-
O <J\ KD G\ 00
R -rT
r- o • • O ro cn (T\ \ £>
V£
cn r-o o cn cn r! cn
I—I CM
o ro cn cn in
V£>
ro OO
CD • O
fd fd •H U fd
-P >
a)
60
The level and variability of these prices suggest that the
second subperiod was very turbulent. The subperiod includes
the volatile years of 1979 and 1980. Prices were relatively
stable during the first subperiod. The largest variance
value is only 1.86 in the six-month forward contract. There
might be an association between price volatility and the
price curve patterns. Prices were generally upward sloping
during the second subperiod, downward sloping during the
first subperiod, and mixed during the third subperiod. It
implies that short-term interest rates are generally higher
than long-term rates during periods of great uncertainty
and volatility. These periods of great volatility usually
occur at interest-rate peaks when the economy is entering
or just about to enter the recessionary phase. Investors
prefer to lengthen the maturity of their debt security
portfolios in order to reap potential capital gain. The
strategy generally results in short rates higher than long
rates at cyclical peaks.
Organized futures exchanges impose a limit on futures
price changes in a particular day. The limit could create a
constraint on futures price movements. However, price
variances of futures and forward contracts do not indicate
any evidence of such a constraint. In many instances, the
futures price variance was greater than the forward price
variance. The nine-month contract provides an example
during the third subperiod, where the futures and forward
61
price variances are 2.5 and 1.91, respectively. In compar-
ing the variances between bond and spot prices, the overall
sample data statistics in Table I show that the bond price
variance was greater for all three contracts. The bond price
variances are 9.7, 9.03 and 7.06, whereas the corresponding
spot price variances are 9.03, 7.06 and 6.47. On average,
price variances decreased as maturity increased.
Since the CIR models' tests require the stationarity
assumption, the T-bill price series is examined for station-
arity. Autocorrelations for the T-bill spot series are
generated for ten lags. The series is nonstationary if the
series has large autocorrelations that decay slowly as lag
increases. If they quickly approach insignificance or are
not significantly different from zero, the series is
stationary. Table II shows the time series is nonstationary
for all maturity periods. The autocorrelations decrease
slowly as lag increases. It is necessary to determine the
minimal degree of differencing so that the tests in this
study do not violate the stationarity assumption. With one
degree of differencing, the serial correlations are insig-
nificantly different from zero at all lags. The correlation
coefficients are mostly within two standard errors (figures
in parentheses equal one standard error). Thus, no further
differencing is needed.
The nonstationarity of the original T-bill spot price
series does not imply the existence of profit opportunities.
62
H
w PQ <
EH
W O H f*i di EH 0 CU CA
h-? H PQ 1
EH
EM
O en o H EH < 1-3 M Pi O O o EH D <
CD (N 1—1 CM rH LO ro N O G\ C\ <Ti in LO o\ Ch •r I -I 0
i—I 1—1
O ro 1—1 LO <r,— •—I CM <D G\ ^
O 00 LO LO ro 00 CM LO 00 00 (T\ 1—1 <y\ rH o 1 00 i—l o o 00 1—I O O LO CM o O r-ro O i—1 r~i • • • • • • • • • • • • • » • •
' ' v—' i "— t v-"' * w ' 1
x—N in in / v rH CM LO 00 r-
CM r- rH in r VO LO O r- 00 o 00 rH (Ts 00 i—i o o 00 i—1 o o <x> CM CM o r-ro CM rH
• • • • • • • • • • • • • • • •
v—' I ' " I ' 1 s*"-' ' ' >—'
LO rH /*—s V£> CM ro u V£> CM in KD ro LO VD CM Ch r-KO CM i—1 0 00 00 rH o o 00 i—1 rH O KD CM O o r-ro O rH u • • • • • • • • • • • • • • • •
u i ^ • — • I I V - / > — • I w I I
*3 00 r- 1 f X—X ro LO ^ s r- 5H LO LO ro in 00 i—1 in 00 LO CT\ o VD rH fd r- 00 i—1 o o 00 i—1 o o KD CM o O 00 ro O i—1 rd • • • • • • 9 • • » • • • • •
c 1 I > — • j N — ' «s_^ f rd •p •
CO r s <T\ ^ i—I KD ro tn r- o m CM m ro 00 ro CM ^ rH a
\ VO 00 I 1 o o (Ts rH O o r CM rH o 00 ro rH rH •H £ • • • • • • • • • • • • • • • • 0 0 * — " I «w | X — ' V • 1 £ -H <D -p SH rd KD s 00 X—N CTi ^ s CM 00 Q) rH ro CM LO CM ro ro in CM CM 00 i> in «H MH CD LO 00 «—1 O O O rH o o r-CM i—1 o 00 CM O rH MH u • • • • • * • • • • » • • • • • •H U • — " r ^ * — —• • — • — • ' — "
0 U U 0 V£> V o> 1—1 / - N r- rH in Q) -P i—1 CM CM in ro CM o in CM CM «—1 00 KD CM -P
C\ i—I o o c\ i—1 o o 00 O i—1 o CT\ CM i—I i—1 MH <! • • • • • • • • » • • t • • » • rd > — • v — ' N W w • —
rd M-! 0 Q u
« - N r- <T\ ,^-x ro KD < u ro ro LO in LO m r- LO CO CM CM KD ^
< 0) ro cr> 1—1 o o o> i—1 O o 00 i—1 o o CM O rH •"3 • • # • • • • • • • • • • • • • tn M 1 — > — • 1 — ' 1 • — ' • — " 1 O •H
O > s ro VD ^ - s CM "sT KD C
LO 00 en m r-00 KO LO CM ro 00 LO 00 i—1 - — s Q)
CM o o o o (T\ o O o rH rH o C7> 1—1 O i—1 U • • • • % • • • • • • • • • • • CD
" — 1 l — > — ' — • > — • W m MH •H r- LO 00 X CM 00 r- ro
r-LO «—i m 00 in KD LO LO 00 1 1 00 r- CM 1—1 Oi o o o a> o i—1 o <T\ o o o <j\ rH O i—1 CD
• • • • • • • • • • • t • • • • U •V—" v—-' v — — • — * ^ — ' — 1 o
UH £ -P >1
Q) £)
0 £ -p • •
•H 0 •H * • • • t • • a> K.
-P -H U £1 w CO w X\ Q fd o 2 -P Xi £1 4 PQ rH -H -P tr> -P -p 4J G 4c Q) M-H fd £ c C C 0 u m g <D 0 Q Q 0 P Q 0 Q a g Q p M Q) S PQ < g m < g PQ < PQ < 0 0 >S
g PQ < CM
< u u £) ro <sO <T\ rH
63
Table III contains the autocorrelation coefficients of T-bill
spot and futures percentage price changes. The coefficients
of spot price changes are low and statistically insignificant
at the first lag. The largest coefficient is only -0.081.
The same pattern is also observed for future price changes
in all maturity contracts. The largest autocorrelation
coefficient of futures price changes is -0.084.
Summary
The CIR models show that the futures and forward prices
differ if future interest rates are stochastic. If these
models are correct, this study should yield the following
results:
(1) If cov(F,B) and [cov(S,B) - var(B)] are signifi-
cantly negative, the futures-forward price differ-
ence should be significantly positive; and
(2) If cov(F,B) and [cov(S,B) - var(B)] are signifi-
cantly positive, the futures-forward price differ-
ence should be significantly negative.
Thus, empirical testing of the CIR hypotheses involves
testing whether cov(F,B), [cov(S,B) - var(B)] and diff(F,f)
are significantly different from zero.
Two inequality equations—var(S - F. ) > var(S - F^ ) 1 2 n m t ,m - n t,n
and var(F . - S, ) > var (F - S )—are derived to test "C. • i f n "c t f in t
whether the T-bill futures market is efficient. The Pitman
64
H H W hi m EH
CO M O a < tc u M U H
W o < Eh 12 W U & w CU CO W & ID Eh P Cm Q £ c Eh O & CO Cm O co 13 o H EH <
kI W & O U O EH
§
CD CM CM 00 LO CM CN N 0>i CT\ <T\ kO CT\ (T\
• H •r rH co CO
v •—s m t£> 00 00 LO cr> 00 00 lO c n vd o CM 00 o 00 LO <7\ KD
o O o O o o O rH o o O o o o • • • • • • • • • • • • • •
\ 1 \
_ rH KD 00 LO r 00 C\ V0 00 CN LO KO
u VD K? LO 00 00 rH o rH LO in o 1 0 un O o o o o o O I—1 O o o o o o
u • • • • • • • • • • • • • •
u 1 V / > — ' I l «•—• — '
w TJ M 05 •—s s s VD r-LO 00 CM 00 KD 00 KD LO rH LO a nT CM CM 00 VD O 00 LO O r- rd O O o o O o rH i—1 o o O o o o -p • • • • • • • • • • » • • •
CO ' — ' l ' — ' 1 ' — ' 1 •>—• 1 r > — '
\ a 0 •H /-s V x—s —». s -P CM LD LO LO rH CM LO LO CO in LO rd 00 00 00 00 o CM LO 00 CM i—i fO •—1 O o O O o O i—1 O O o o o o <L> • • • • • • • • • • • • • t u
1 — ' 1 >•—• ^ — ' •*— • — -
u
0
u 0 •P /•—"V <•—«» <—V s—N. ,—s 3 CM in 00 LO rH CM VD CN OS m 00 LO < i—1 CM 00 CM o LO LO CM «s? rH <vT
CN O o o o O O rH rH o o O o O O <+H • • • • • • • • • • # • • •
0 1 — — 1 1 w
i 1 5-1 0 TJ
/ — s X-—N ^ - - S / ~ V
O LO LO LO G\ CN rH 00 LO <N rH LO lo KD vo 00 00 00 o rH LO I—1 00 i—i O o o o O O O rH O O O o o o • • • • • • • • • • • • • •
1 1 i 1 1 --
U) 0) cn Q) a) 0 •P U u u
>i -P -P •P 0 0 0
-P 1 ^ o 0 0 QA -P -p •p •H -P a 0. 04 CO 3 2
U tn CO CO CO UA Cm 3 : £ Cm
+J (1) X i •G rG •P A rC
rd -p -p -P G 4J -P -P kr-H c a £ 0 C a a 0 0 0 g 1 0 0 0 g a g
g 1 g g g 1 i 1 CN 1 i i ro V£> a \ rH 00 vo <T\
65
statistic provides the significant testing of the inequali-
ties. Regression tests of market efficiency involve testing
whether the slope coefficient is significantly different
from zero.
Testing the CIR models and the T-bill futures market
efficiency hypothesis requires the estimation of average
covariance and variance. The local covariances and
variances over each contract period in the sample are com-
puted and they are then averaged over all contract periods.
The mean of diff(F,f) is estimated from the weekly futures-
forward price difference data.
This study is confined to analyzing the first three
T-bill futures contracts. Weekly data have been collected
from the Wall Street Journal. The sample period begins on
March, 1976 and runs through December, 1984. The division
of the overall sample period into three subperiods permits
an examination of the sample data behavior. The stationarity
of the relevant price series is also examined.
CHAPTER BIBLIOGRAPHY
Huang, Roger, "Efficiency and the Volatility of Flexible Exchange Rates," unpublished manuscript, University of Pennsylvania, 1980.
Roll, Richard, Behavior of Interest Rates, New York, Basic Books, 1970.
Snedecor, George and William Cochran, Statistical Methods, Ames, Iowa, Iowa State University Press. 1980.
66
CHAPTER V
RESULTS
This chapter presents and interprets the results of
the statistical tests to determine whether they are con-
sistent with the hypothesized theory. The first section
gives the Cox-Ingersoll-Ross models' test results and
interpretation. The next section reports the findings and
discusses the CIR models' ability to discriminate among the
price differences. Subsequent sections cover the results
from the variance and regression tests of market efficiency.
Tests of the Cox-Ingersoll-Ross Propositions
Table IV summarizes the results of the tests of Eqs.
[5] and [6]. (All covariances and variances are magnified
by 100 times to reduce the large number of decimal places.)
The results for Eq. [5] indicate that the mean futures-
forward price differences in all contracts are significantly
different from zero. On average, futures prices in the
three- and six-month contracts are respectively 0.31 and
0.12 greater than the corresponding forward prices. How-
ever, forward prices exceed futures prices by 0.28 in the
nine-month contract. The null hypothesis of an insignificant
mean price difference is, therefore, rejected for all three
67
68
> H
£1 A PQ < Eh
co A w Q o g
& H
u
fa o co Eh h3 D co w
Eh CO fa Eh
V£>
C 0 •H -P ftf 3 tJ1 W
LD
c o •H +3 fd 3 &1
w
ffl
CO CM •
> rH 0 o
M—!
fa
UH UH •H Q
MH
fa
MH M-l •H Q
rCj i—1 > i - P fd tr> 0 C
•H Q) 0) - P rH i-q CO X)
•H fd > i •P •H - P fd U •H - P fd CO > 2
- P fd
g
i> LO V£> •NT <sO o o CM o • • • • • •
2 5 CN ro 2 5
ro ro
«H G\ H h in h* • • • i—i^I—i r^
CM
ro
ro
vo H r-• • •
CN
00 ro ro r • • •
oroHCTt
r-H ro to ro KD KO • • •
^ H <n r-
CM in 00 KO m rH ro <sO CM CN VD
0,
3,
i—1 CM 0. • L
i.
2
r-ro
m
fa •
• — o > 0 CJ
CD 3 rH fd
>
• fd • p o rd -h
C-P4J (0 rd CO -H N CD I M -H S - P O W
f i r
LD CM
ro
ro in ID h h • • •
H r o H ID ro
r -r- o r-
c h h CM
rH ro IT) CM m 00 KD LO ro • V£> • KD H ro VD CM CM V£>
o (T\ rH <T\ o ro •—1 CM o • •
I—1 CN r- in ro CM
I I I
£1 -P a 0 g
1 VD
-P fd
a -p fd co a) i
g -P
CD 3 H fd
>
i—J fd o *H -P •H U CO
+3 C 0 g 1
<y\
-P fd
a -p rd co a) r
g -p
a) rH fd
>
fd o •H -P <U •H N JH -H CJ CO
69
contracts. The null hypothesis of an insignificant mean
local covariance between the futures and bond percentage
price changes is also rejected for all futures contracts.
The mean local covariances are 0.99, 1.64 and 0.77.
A comparison between the mean price difference and the
mean covariance for the three-month contract reveals no
support for the CIR proposition embodied in Eq. [5]. With a
positive covariance of 0.99, the mean price difference should
be negative. The summary statistics for the six-month con-
tract also indicate no support for the proposition. However,
the nine-month contract's results are consistent with the
CIR models' prediction. With a positive mean covariance of
0.77, the mean price difference is negative as hypothesized.
There is little evidence to suggest that the CIR model
explains why the futures and forward prices differ.
There is more support for the CIR proposition embodied
in Eq. [6]. Table IV again shows that all mean price differ-
ences, covariances and variances are significantly different
from zero. It also shows the mean price values are an
inverse function of length to contract maturity. The mean
values of 6.31, 1.2 and 4.6 in the three-month contract
decrease to -0.28, 0.36 and 1.11, respectively, in the nine-
month contract. The figures for the first two contracts
show support for the CIR proposition in Eq, [6], With the
mean variances greater than the mean covariances, the mean
70
price differences are positive. The results are consistent
with the model's prediction. However, the mean price differ-
ence of -0.28 in the nine-month contract is inconsistent
with what Eq. [6] would predict.
The evidence in Table XV is mixed. If the underlying
CIR proposition were valid, the empirical test results of
Eqs. [5] and [6] should have led to similar conclusions.
Both equations express the same proposition—but in differ-
ent ways. The evidence in this study is inconsistent and
contradictory.
There are several possible explanations for the incon-
sistent test results. One possible reason is the invalidity
of the CIR models. The futures-forward price differences
may not be attributed to the daily settlement process and
stochastic interest rates. The price differences might not
be related to market participants' expectations of the rele-
vant covariances and variances. Other factors may cause the
price discrepancy.
A second possibility is that the CIR models are valid
only for non-financial underlying assets. The first CIR
model indicates that the futures-bond covariance should be
negative in order to obtain a positive futures-forward price
difference. This is highly improbable for financial assets.
Since this study deals with T-bills and it specifies that
the bond price is the present spot price of a T-bill
71
expiring on the relevant futures contract's maturity date,
the futures and bond prices would always exhibit a positive
covariance. Prices of financial spot and futures assets
move inversely to interest rates. Accordingly, the futures-
forward price difference should be negative. The possibility
of a positive price difference is remote unless there is a
basic change in the economic relationship between financial
assets' prices and interest rates. The point is that the
CIR models may not adequately describe the price behavior of
financial assets and their corresponding futures contracts.
The inconsistency of the test results may also be
attributed to measurement errors. There is no explicit
forward market for T-bills. The implicit forward prices may
not accurately represent such prices. Factors that affect
explicit prices could be completely ignored in determining
implicit prices. Moreover, spot prices that are used in
computing implicit forward prices may introduce further
measurement errors. Because of the matching problem between
futures contracts and the relevant T-bills, it is necessary
to use the nearest spot prices as substitutes. The problem
is especially acute in determining the implicit nine-month
forward contract prices.
Intra-Sample Variations Among Futures-Forward Price
Differences
The ability of Eq. [5] to explain futures-forward price
differences is questionable. The results in Table V provide
72
w
EH
CO w u a
w fa
fa H Q
fa 0 H &
5
1 0 fa 1
CO fa 6 H) & D fa
fa O
co !7r
Eh < H & < > W P3 &
CO I
53 H
2 0 H
0 •K i n r - r o
<—X rd rd rd rd rd rd rd rd rO rH i—1 w CD • • • • s— > d d d d rH d d d d o o o 4
2
5
M •H 1 1 1 rd - P > •H 1 W
0 PQ
K. co >—*
> & 0 3 O 0
U O LO
i—1 00 cm r o vo m r o <d (1) c o m O rH r o o o CM i—1 rH > • • • * • * • • • • • •
- H o o CM o o rH CM r - o o CM CM - P r o CM i—1 rd I 1 1 t P CD 53
CU 2 0
o i n 03 m CM KD
<U oo CM CM r o KD o> O i n > r o r o CT> o rH r o 00 O CM O H
•H • • • • • • • • • • • •
- P o o CM H o o 1—I o j r - o o r o CM oo •H r o CM rH CO f 1 1 0
x—, o» m
fa —' cu > 2 0 0 u U
0 i n r -
0 I > o 00 > O i—1 rH rd rd rd rd rd rd rd rd
•H • # • •
•P o o i—1 r o d d d d o d d d d rH i rd
d d rH i
tr> 0) 23
I
r d d CD d CD d CD •P O 3 0 2 0 2
H > i t P •H r—j •H rH •rH i—1 rd - P rd - P rd - P rd o CD rd > rd > rd >
•H <D J •H •rH •H •P rH > i—I > i—1 > i—1 CO ^ > 1 <D • rd 0 • rd CD • rd
*h rd - p Q o Q +> O Q •P o - P -H •H JG rd •H rd •rH ^d rd *H rd u u • p d • +> - P Q) +J d • - P - P Q) - p d • - P - P 0 j - P rd 2 d rd rtf 00 •H N d rd n 3 CO •H N d rd CO - H N CO > - P 0 <D +3 1 &H "H 0 Q) - P 1 ^ -rH 0 CD - P 1 M *H
td *>• ^ g CO 1
- P U CO g g co 1 •P U CO g g CO | - P U CO
r o 1
VD I
o\
CD
rd U rH H Oi Pu rd
rd
73
no evidence that the futures "bond, cova.iria.nce is useful in
discriminating among the futures-forward price differences.
With positive covariances, the average price difference in
the three-month contract (third column) should be negative.
However, the observed average price difference is 0.34 and
significantly positive. For the group with negative
covariances, the average price difference in the three—month
contract is 0.075 and insignificant (only four observations).
For the six-month contract, none of the observations has
negative covariance. For those assigned to the group with
positive covariances, the average price difference is 0.122
and insignificant. The only average price difference that
is significant and has the right sign is that of the nine-
month contract in the positive covariance group. The mean
price difference is -0.292.
Table V also shows that Eq. [6] is not useful in
explaining the intra-sample variations. Observations with
smaller spot-bond covariances than bond variances are
assigned to the second column. The rest are assigned to the
third column. If the C1R model proves useful, the average
futures-forward price difference in the second and third
columns should be positive and negative, respectively. The
average price difference in the three—month contract (second
column) is 0.31 and significant. However, the price differ-
ences of 0.12 and -0.26 in the six-month and nine-month
contracts are not different from zero. No result is
74
obtained for the group with larger spot-bond covariances
than bond variances because of inadequate data.
The results in Table V do not present a consistent
pattern across all contracts. The results reject the CIR
models. The models do not seem able to discriminate among
the price differences. The models' predictions could be
more accurate with better proxies of the market expectations
of the relevant variances and covariances. This study uses
ex—post variances and covariances as proxies for market
expectations. The ex-post estimates of expectations may
contain measurement errors. Moreover, the estimates are not
obtained from observations across the whole six-month and
nine-month contract periods. To avoid overlapping estima-
tion periods, this study uses data from only the first three
months of the six-month and nine-month contracts. Thus, this
study implicitly assumes that the variance and covariance
are constant across the relevant contract period. This
assumption may not hold. These measurement errors could
contribute to the inconsistent results reported in Table V.
Variance Tests of Market Efficiency
The variance tests indicate that the futures market is
efficient. Tables VI and VII present the test results for
Eqs. [20] and [23], respectively. Table VI shows that the
mean values of Var(Sm - F ^ ) are greater than those of
Var(S - F ) for all t's in the first set of variance n t,n
75
H > w a CQ <
o CM
53 O H Eh <c o a w fa o (/} EH £>
fa PS EH 0} fa EH fa O § H P$ c >
Cri
tical
Val
ue
Cri
tical
Val
ue ro ro ro ro ro • • • « •
o o o o o ro co ro ro ro • • • • • o o o o o
ro ro ro ro ro » • • t • o o o o o
Pit
man
cni in co h ^ If) 00 <T\ O • • • • • O O O O rH
ro oo vo rH ro in ro ro • • • • • o o o o o 1
m H in oo CN o> o • • • • • O O C O rH
Co
rr. G\ ro CM 00 in (N o ro • • • »•
o o o o o
00 H VD H v£5 m ko m ko • • • • • o o o o o
in rH rH ^ CM CM CM O rH o o o o o
Siz
e to in in m in ro ro ro ro ro m m in m in co ro co co ro
m in in in in ro ro ro ro ro
* *
Var
(2)
*
H (N h hin ro ro ro kd cm • » • • • in ro rH o o
v£> oo cm o\ KD 00 CM rH O • • • • • Mn m m VD
H CN LO ro co ro vo cm • • • • * in ro H o o
Siz
e in in m in m ro ro ro ro ro ^ ^ ^ ^ ro ro ro ro ro ^ ^ ^ ro co ro ro ro
Var
(1
)**
r- KD 00 CN o kD 00 CM H C • • • • • Mn in in vo
ro r- oo ^ oo oo r- CM rH rH • • • • • kD Is- 00 h 00
ro r- oo oo 00 CM i—1 i—! • • • • • KD 00 r- 00
-P H r- o cm i—1 i—I H ^ r- O CM 1—1 1—1 rH ^ r- O CN i—1 rH
|
Set
*
Fir
st
Sec
ond
Th
ird
<r» ro
U •H -C3 •
-P "N £ n3 -
a •P fd fa
t • K a VD C/l CN -—•
u II fd > £
CM ro ' — '
u II fd II > e *
* ** *
-P CD 10 'O c 0 0 CD cn • ro rH II • a g
V£> +>
CM fa
II 1 £ en E • — -
u fd -P >
CD W -P i—i V)
v—• u u •H fd fa • > * ro •K
i—1 *
II a
76
comparisons, where ni — 26 weeks and. n — 13 weeks. The case
of t = 1 exemplifies the findings where Var(Sm - and
VarfS - F ) are 7.67 and 5.31, respectively. The Pitman n t ,n
statistics indicate that all variance differences are
significant; hence the inability to reject the null hypothe-
sis for all t's. The table also shows similar findings from
the other two sets of variance comparisons, except for two
cases. In the second set where t = 1, Var(Sm - F
t^ m) equals
6.83 and is less than Var(S„ - F ). In the third set n l. f ii
where t = 1, the inequality is insignificant—the computed
Pitman statistic is only 0.25.
In general, the variance value is directly related to
length to maturity. The variances are largest where m = 39
weeks and are computed as 6.83, 7.77, 8.28, 7.14 and 8.18.
In contrast, where m = 13 weeks, the corresponding variances
are 5.31, 3.32, 1.37, 0.67 and 0.25. The variables, for
which the variances are computed, are forecast errors. The
error terms become more variable as maturity lengthens.
The results also indicate that the correlations between
the variables in the first set are smaller than those in the
second set. The forecast errors of longer-maturity con-
tracts are probably more closely related because information
differences between them is less than the difference between
nearby contracts. Error terms are least correlated in the
third set between contracts in the third set where m = 39
weeks and n = 13 weeks.
77
H H >
W
CQ <
EH
ro CM
53 O H EH < £3 a W
0
w
EH
D m
* 3 &
EH
W EH
W U
1 H &
C >
fd a) O ^ •H rH -P fd • H > M U
fd e -p
•rH PU
N
O O
A) N
- H cn
* *
CM
U fd >
CD N
• H CO
U
'D >
CM CM CM CM KD VO KD
ro ro ro ro CO ro ro CO ro CO ro • • • • • • • • • • •
o o o o o o o o o o o o
LO co R - oo CT\ <J\ CT\
LO 00 00 rH oo r-
lo ro rH oo ro oo r-
O O O O O O O O O O O O
CN
VO *—I o ro I—1 rH » * • •
O O O O
CM O CO CM RO CM CTI LO RO M T ) H
o o o o o o o o
LD in LO ro ro ro ro
R - R - R -CM CN CM CM
^ I N M rH CM CM CM
Hi H fd
[CM
J4 fd
ro H oo rH rH rH O • • • •
O O O O
LO CM ro LO ro CM ro
m ^ co
KD KD LO KD
O O O O O O O O
LO M LO ro ro ro ro
R - R - R - R -CM CM CM CM rH CM CN CM C/}
\
CM VD rH CM 00 ^3 0> CO • • • •
O rH O O
oo a\ ^ IT) CO • • • •
O O O O
H CM ^ M CM CM
• • • •
O I—I I—1 I—I
-H 4--p
PM
u
fd >
O CO VD O r O ^ D O ^ k O ro VD CTi
w tfl cn M M
CD CD CD CD CD CD S £ s
ro KD CPs
i—1 CM ro
u
fd >
*
78
The test results for Eq. [23] in Table VII reinforce
the findings from Table VI—the futures market efficiency
hypothesis (null hypothesis) cannot be rejected. The values
of Var(F,,• n - S ) exceed those of Var(F - Sfc), except t+ifn
f
for t = 0 in the six- and nine-month contracts. The case
of t = 3 i n the three-month contract provides an example,
where Var(Ft+i>n - Sfc) and Var(F^ n - Sfc) equal 1.66 and
0.19, respectively. The table also shows that the correla-
tion coefficients are also a direct function of length to
maturity on an inter-contract basis. The correlation
coefficients of 0.33, 0.72, 0.59 and 0.75 in the nine-month
contract are greater than those in the three- or six-month
contracts. In Eq. [20], the correlation is computed between
forecast errors in different contracts. In Eq. [23],
however, the correlation is computed between price differ-
ences at two different time points (four weeks apart) m the
same contract. The lack of information for contracts farther
from maturity may be a possible cause for the greater cor-
relation. Market participants are less able to distinctly
distinguish between the prices at two time points for longer-
maturity contracts. The difference between prices becomes
less obvious to them due to less information. The market
sees the prices at different time points becoming more
related for longer~roa.turity contracts.
79
Regression Tests of Market Efficiency
Before examining the regression results, residuals are
tested to determine the regression model's appropriateness
for the data. An effective way to examine a set of residuals
is through graphic analysis. Residuals are plotted against
the relevant independent variables for all contracts and t's.
All the residual plots (fifteen of them) show no presence of
any systematic trends. There is no evidence of error variance
nonhomogeneity or nonlinearity. To determine whether the
data are substantially non-normal, this study generates
residual normal probability plots for all contracts and t's.
The evidence from the skewness and kurtosis coefficients is
mixed, but the plots show no indication of serious departures
from normality.
The residual analysis indicates no serious violation of
regression assumptions—homocedasticity, normality, and
linearity. No data transformations are necessary. The
residual independence assumption is examined during the
review of the regression results•
The regression results (Table VIII) indicate that the
null hypothesis (b=l) cannot be rejected for the first four
t's associated with the three-month futures contract. The
t statistics (in parentheses) of -1,64, -0.7, -0.29 and
-0.21 are less than the critical value. However, the slope
coefficient for t = 12 weeks is not randomly distributed
30
H H H >
W
CO <
EH
0} EH J p w w c*
En cn W Eh
S 3
O H 00 CO W &
a w &
r H
td O
•H -p •H rH u cd a >
Cm
cd O
*H (D -P 3 •H rH u cd o >
CD N
- H cn
5 p
PM
CD
04 o I—I 03
-p Q* CD O 5H CD -P C
15 LO
rH 15
15
15
,15
.15
.15
.15
.15
.15
.15
• • • • •
o O O O o • • • • •
CM CM CN CM CM
O O O o O O •
CM CM CM CM CM CM
LO ro
LO ro
LO ro
in ro
LO ro
LO ro
LO ro
LO ro
in ro
LO ro
i—I
rH
CM
CO i—1 CM CO CO •
(N CM CM rH rH rH
rH ro
ro
ro VD
oo O
KD ro
ro
ro
CM CM
r-CM
•
tJ* cr» VD KD ro 0} VJD 00 OS I • • • •
o O o o
00 as
o
CM
CO ro
cr» <Ti 00 Ch . • •
•—i rH o O
CM CM 00 o • • • •
C\ CO ro 00 CM CM rH
r- ON in r- «—i CM
CM ro ro CM • t • . . . .
o o o o o o O
o ^ ( N ^ o o c n o ^ « H ^ ^ r^^DCTir^ONCMcrkCMO^^ # » » » • • • * • •
O i H O O O O O O O C N I I I I I
rH^-r-oo to 00 r- ro ^ ^ D C N r ^ r o o o r o o o r o r ^ o v o o o k o o • • • • • • • • • • . . . .
O CM O i—1 O i—I O t—I O (N O rH O CM I I I ! * 1 »
r^<£>l>rHV£>V0CM<^rHCM r - v D ^ r ^ c M r o o o r H ^ J 4 ^ . . . . . • • • • •
O H h O C N O O C l / l C M
CM
LO
ro
,—» /-N ^ - s >
in 1—1 r^ CO CM KD
in ro CM ro CO ro 00 1—1 00 00 o cr> • • • . . . . • . . . .
o rH 00 i—1 00 rH r- CM (T\ 1—1 KD i—I CM rH — 1—1
w CM CM ' ro '
CM
w M CD 0 5
co
10
CD CD &
CM
in X CD CD &
o> co
81
*3 CD P C -H 4-> a 0 u 1 i
H H H >
w J CQ C EH
rH fd u
•H CD -P 2 •rH rH
u (d u >
En
fd O •h a) +3 3 •rH r—1 M rd u >
CD N •H i n
p
P>4
0) P.* O i—I w
• • • • •
m LO in C—1 rH i— i • • • •
• • •
• • • * •
04
04
.04
.04
.04
.04
.04
.04
•
CM •
CM CM CM CM CM CM CM
ro ro on ro ro ro ro "sP ro
rH r o CM • • • * •
• *
i—t i—l r—1 • * * • •
rH i n * • •- • • • •
r - CM a s rH • • •
• •
tr1 r - ro CO rH CM CM
i * • •
o o o
ld in m ^ LO rH LO CM VO oo in in • • • • • •
O CM O rH O CM I I I
c \ ( N [ ^ H o o i n c n ^ c M i i i ^ • ^ c M o o o o o c M C M C M r -• • • • • • • • • •
o c N i o o o o r o o r o o r o I I 1 I I
a. O LO CD "sT H CT u • • •
SH O CM rH A> ^ ^ ro -P A H
/ s in < J \ N
CTs 00 LD CM in r- 00 00 CM CTs
• • • • • • • • • • • * •
rH o CM K D CM VD CM ro ro ro O ro < D 00 s—' VD
v— r - "" '
CM O rH
CM
CO * CO
M M CD • CD CD -P CD 5 £ S
0 <Y> O O ro ro
O O O -P 2 fd
u o JMH
CD -P CO 3 n <
* *
10 o •H •P CO •H •P rd -P CO •p
•K
82
about one. The intercept terms are not significantly
different from zero, except for t = 12 where the t statistic
is 2.42. The R-squares are highly significant and increase
as the futures contracts approach maturity. The R-square
for t = 1 is 0.49, but it increases to 0.98 for t = 12. As
length to maturity decreases, more information is available.
It leads to better estimates of future spot prices, hence
greater R-squares. Table VIII also indicates no serious
autocorrelation problem with the three-month contract. The
significance level for the DW statistics is 5 percent.
The results for the six-month futures contract indicate
that the null hypothesis cannot be rejected for t = 4, 7 and
10 weeks. The relevant t statistics are -1.38, -1.3 and
-1.35. The intercept terms for the above t's are distributed
randomly about zero. The R-square values are smaller, but
they follow a similar pattern observed in the three-month
contract. Some autocorrelation exists for the one and four
week regressions where the DW statistics are both 1.3. It
is minor, however, and should not seriously affect the
regression estimates or the related statistical tests.
It appears that the null hypothesis cannot be rejected
for t = 1, 4 and 10 weeks in the nine-month futures con-
tract, with the relevant t statistics of -1.83, -2 and -1.84.
The intercept coefficients for the above t s are not
significantly different from zero. However, Table VIII shows
that substantial serial correlation exists for all t s in the
83
nine-month contract. Consequently, the parameter estimates
are not efficient, although still unbiased. More seriously,
the standard errors are underestimated. Thus, it understates
the ability to reject the null hypothesis. To adjust for
this problem, re-estimation is done using Yule-Walker (1)
equations. The adjusted estimates indicate the slope coef-
ficients for all t's are significantly different from one.
The relevant t statistics are -2.42, -2.81, -3.85, -3.2 and
-3.75. The null hypothesis is rejected at the 5 percent
level.
There are several possible reasons why serial correla-
tion exists in the nine-month contract. There could be
systematic measurement errors in the relevant variables.
This is not a strong argument in this case because the
market is generally efficient for the three- and six-month
contracts. The market should be inefficient throughout,
unless the errors are only in the nine-month contract data.
Another reason could be the omission of a significant
variable related to information costs. More information
and a longer learning process are necessary to obtain longer-
term forecasts. The variable becomes relevant for longer-
maturity contracts—that may explain the existence of auto
correlation for all t's in the nine-month contract and for
t = 1 and 4 in the six-month contract.
A third possible, but unlikely, explanation might be
default risk. This might also be relevant for longer-maturity
84
futures contracts. The underlying asset is default-free,
but the futures contract is not. Although organized
futures exchanges could promise settlement, future economic
conditions could dictate otherwise. Defaults on futures
contracts have happened. Prices of longer-maturity futures
contracts may contain default risk premiums, leading to
serial correlation in the residuals.
The R—square values in the nine—month contract do not
increase monotonically as observed in the three- and six-
month contracts. It implies that more information is not
readily available to the market in making better estimates
as length to maturity decreases from thirty-nine to twenty-
seven weeks. The longer-term forecasts may require more
information than the passage of time allows during that
period (thirty-nine to twenty-seven weeks before maturity).
Summary
This chapter presents three sets of results and their
accompanying interpretation. There is not much support for
the two CIR models. Although there is a little more support
for the second model, it is not known why the test results
from the first CIR model contradict those from the second
model--both are derived from the same underlying hypothesis.
The models' inability to explain intra-sample variations
among futures-forward price differences further shows the
lack of support for the CIR models.
85
The second and third sets of results pertain to the
efficient market hypothesis. The variance tests indicate
that the T-bill futures market is highly efficient. The
regression test results generally support the efficient
market hypothesis.
CHAPTER BIBLIOGRAPHY
1. SAS Institute, Inc., SAS/ETS User's Guide, 1982 ed., Cary, North Carolina, SAS Institute, Inc., 1982
86
CHAPTER VI
SUMMARY AND CONCLUSION
This study examines two aspects of the T-bill futures
market. First, it tests whether two Cox-Ingersoil-Ross
models might explain price differences between futures and
implied forward contracts. CIR hypothesize that stochastic
interest rates contribute to the futures-forward price
differences. The effect of interest rate stochasticity on
the futures—forward differential is a result of the daily
settlement process.
This study also investigates the efficiency of the
futures market using variance and regression tests. The
variance tests investigate whether futures prices are too
erratic for the market to be efficient. The regression
tests examine whether futures prices are unbiased predictors
of future spot prices.
Expressing the futures-forward price difference in
terms of futures and bond prices, the first CIR model
implies that if the local covariance between the futures
percentage price changes and bond percentage price changes
is positive, the futures-forward price difference should be
negative. If the covariance is negative, the model predicts
87
88
a positive futures-forward price difference. The test
results provide little support for the model.
The second CIR model expresses the same hypothesis in
terms of spot and bond prices. If the local bond variance
of the percentage price changes is larger than the local
covariance between the spot and bond percentage price
changes, the model predicts a positive futures-forward price
difference. If the variance is smaller, it predicts a nega-
tive price difference. Test results do not support the
model. The results cast doubt on the validity of both
models. The models could be correct, with the findings
being due to measurement errors or futures market ineffi
ciency. Futures market efficiency is the second issue in
this study.
Efficiency in the T—bill futures market is stated in
terms of a fair game. On average, the futures price should
equal the expected spot price; i.e., the forecast error is
orthogonal to any element of the information set. The
futures price should vary only with the random arrival of
new information. If the futures price movements are con-
sistently excessive relative to actual subsequent events,
the market is inefficient. The inefficiency may be
attributable to either futures prices not reflecting all
available information, or the market rationally and
persistently expecting events that do not materialize.
89
Krasker (1) refers to the latter as the "Peso problem.' The
market is efficient, but it may be concerned with an adverse
effect that does not happen. Although the probability of
the event may be low, the market rightfully reflects it
period after period. Even if the event has not yet happened,
it may and should, therefore, be impounded in prices.
The theoretical basis for variance testing of market
efficiency is that the forecast variance should be lower
than the realized variance. The tests require that the
relevant time series be stationary. Using two inequality
equations, this study obtains results that strongly indicate
the T-bill futures market is efficient for all three con-
tracts. Market participants take into consideration all
available information in generating their forecasts (futures
prices).
The regression test equation is derived from the same
basic model that generates the variance test equations. The
T-bill futures price should be an unbiased predictor of the
future spot price if the market is efficient. The regres-
sion tests posit a slope coefficient of one for an efficient
market. The test results indicate the efficient market
hypothesis cannot be rejected for the three- and six-month
contracts. They also show that the market for the nine-
month contract is inconsistent with market efficiency. The
existence of autocorrelation might be due to either measure-
ment errors, omission of a time—related variable, or a
90
default-risk factor. The regression tests could be sensi-
tive to one (or more) of these factors while the variance
tests are not. That would explain the difference in results
for the nine-month contract.
The R-square values increase as the three- and six-
month contracts approach maturity. Although the R—squares
in the nine—month contract do not increase monotonically,
they are less than those in the three— and six—month con-
tracts. As time passes, better information is available
through accumulation. The market obtains better estimates
with better-quality information. A lack of good quality
information does not imply inefficiency as long as prices
impound all available information existing then.
Implications for Further Research
This study's results suggest several avenues for fur-
ther research. This study tests the CIR models in the
T—bill futures market. French has done work in the copper
and silver futures markets. To date, the evidence is mixed.
Testing other futures markets would provide further evidence
of the CIR models' validity.
This study suggests work in futures markets that have
corresponding explicit forward markets. It eliminates the
necessity of generating implicit forward prices that may
not be representative.
91
Regression tests in this study examine whether futures
prices are effective and unbiased predictors of future spot
prices. A possible avenue for further research is examining
whether implied forward prices provide unbiased estimates of
future spot prices. A related area for further exploration
is determining which prices (futures versus implied forward)
are better spot price estimates.
Another possible area for further research is deter-
mining why some regression and variance test results differ.
Although both types of tests are derived from the same
underlying model, one could be more powerful than the other.
This study uses data in the first thirteen weeks of
any contract period to avoid overlapping data. It may be
desirable to expand the sample period so that adequate non-
overlapping data can be obtained. Since futures trading in
T-bills started in 1976, it could be several years before
enough non-overlapping data can be obtained for an adequate
sample size.
Conclusion
The results of this study support two conclusions;
(1) CXR do not adequately explain futures price structure,
and (2) the T-bill futures market does appear to be effi-
cient even though the prices do not behave as CIR would
expect.
CHAPTER BIBLIOGRAPHY
1. Krasker, William, "The 'Peso Problem" in Testing the the Efficiency of Forward Exchange Markets," Journal of Monetary Economics, VI (April, 1980) , 269-276.
92
BIBLIOGRAPHY
Books
Chatfield, Christopher, The Analysis of Time Series: Theory and Practice, New York, Halsted Press, 1975.
Harvey, Andrew, Time Series Models, New York, Halsted Press, 1981.
Kleinbaum, David and Lawrence Kupper, Applied Regression Analysis and Other Multivariate Methods, North Scituate, Massachusetts, Duxbury Press, 1978.
Kolb, Robert, Interest Rate Futures: A Comprehensive Intro-duction, Richmond, Virginia, Robert F. Dame, Inc., 1982.
Mason, Robert, Statistical Techniques in Business and Economics, Homewood, Illinois, Richard D. Irwin, Inc., 1978.
Roberts, Harry, Conversational Statistics, Palo Alto, California, The Scientific Press, 1974.
Roll, Richard, Behavior of Interest Rates, New York, Basic Books, 1970.
SAS Institute, Inc., SAS/ETS User's Guide, 1982 ed., Cary, North Carolina, SAS Institute, Inc., 1982.
Snedecor, George and William Cochran, Statistical Methods, Ames, Iowa, Iowa State University Press, 1980.
Wesolowsky, George, Multiple Regression and Analysis of Variance, New York, John Wiley and Sons, 1976.
Articles
Arak, Marcelle, "The Effect of the Tax Treatment of Treasury Bill Futures on Their Rates," Journal of Futures Markets, III (Spring, 1983), 65-73.
Black, Fischer, "The Pricing of Commodity Contracts," Journal of Financial Economics, III (January/March, 1976), 167-179.
93
94
Bigman, David, David Goldfarb, and Edna Schechtman, "Futures Market Efficiency and the Time Content of the Informa-tion Sets," Journal of Futures Markets, III (Fall, 1983), 321-334.
Burger, Albert, Richard Lang and Robert Rasche, "The Treasury Bill Futures Market and Market Expectations of Interest Rates," Federal Reserve Bank of St. Louis Review, LIX (June, 1977), 2-9.
Capozza, Dennis and Bradford Cornell, "Treasury Bill Pricing in the Spot and Futures Markets," Review of Economics and Statistics, LXI (November, 1979), 513-520.
Cargill, Thomas and Gordon Rausser, "Temporal Price Behavior in Commodity Futures Markets," Journal of Finance, XXX (September, 1975), 1043-1053.
Chow, Brian and David Brophy, "Treasury-Bill Futures Market: A Formulation and Interpretation," Journal of Futures Markets, II (Spring, 1982), 25-47.
Cornell, Bradford, "A Note on Taxes and the Pricing of Treasury Bill Futures Contracts," Journal of Finance, XXXVI (June, 1981), 1169-1176.
, "The Relationship Between Volume and Price Variability in Futures Markets," Journal of Futures Markets, I (Fall, 1981), 303-316.
Cornell, Bradford and Marc Reinganum, "Forward and Futures Prices: Evidence from the Foreign Exchange Markets," Journal of Finance, XXXVI (December, 1981), 1035-1045.
Cox, John, Johathan Ingersoll and Stephen Ross, "The Rela-tion Between Forward Prices and Futures Prices," Journal of Financial Economics, IX (December, 1981), 321-346.
Dusak, Katherine, "Futures Trading and Investor Returns: An Investigation of Commodity Market Risk Premiums," Journal of Political Economy, LXXXI (November/December, 1973), 1387-1406.
Emery, John and Robert Scott, "Evidence on Expected Yields Implied from the Term Structure and the Futures Market," Business Economics, XIV (May, 1979), 22-27.
Fama, Eugene, "Efficient Capital Markets: A Review of Theory and Empirical Work," Journal of Finance, XXV (May, 1970), 383-415.
95
Flavin, Marjorie, "Excess Volatility in the Financial ^ Markets: A Reassessment of the Empirical Evidence, Journal of Political Economy, XCI (December, 1983), 929-342.
French, Kenneth, "A Comparison of Futures and Forward Prices, Journal of Financial Economics, XII (November, 1^»3), 311-342.
Friedman, Benjamin, "Optimal Expectations and the Extreme Information Assumptions of 'Rational Expectations Macro-Models," Journal of Monetary Economics, V (January, 1979), 23-41.
Grossman, Sanford and Robert Shiller, "The Determinants of the Variability of Stock Market Prices," American Economic Review, LXXI (May, 1981), 222-227.
Howard, Charles, "Are T-Bill Futures Good Forecasters of Interest Rates?," Journal of Futures Markets, I (Winter, 1982), 305-315.
Huang, Roger, "The Monetary Approach to Exchange Rate in an Efficient Foreign Exchange Market: Tests Based on Volatility," Journal of Finance, XXXVI (March, 1981), 31-41.
Jacobs, Rodney and Robert Jones, "The Treasury Bill Futures Markets," Journal of Political Economy, LXXXVIII (August, 1980), 699-721.
Jarrow, Robert and George Oldfield, "Forward Contracts and Futures Contracts," Journal of Financial Economics, IX (December, 1981), 373-382.
Kane, Edward, "Market Incompleteness and Divergences Between Forward and Futures Interest Rates," Journal of Finance, XXXV (May, 1980), 221-234.
Kolb, Robert and Gerald Gay, "The Performance of Five Cattle Futures and Predictors of Subsequent Spot Prices, Journal of Futures Markets, III (Spring, 1983), 55-63.
Koppenhaver, G. D., "The Forward Pricing Efficiency of the Live Cattle Futures Market," Journal of Futures Markets, III (Fall, 1983), 307-319.
Krasker, William, "The 'Peso Problem' in Testing the Efficiency of Forward Exchange Markets," Journal of Monetary Economics, VI (April, 1980), 269—276.
96
Tana Richard and Robert Rasche, "A Comparison of Yields on Futures Contracts and Implied Forward Rates," Federal Reserve Bank of St^ Louis Review, LX (December, 1978), 21-30.
LeRoy, Stephen and C. J. LaCivita, "Risk Aversion and the Dispersion of Asset Prices," Journal of Business, LIV (October, 1931), 535-547.
LeRoy, Stephen and Richard Porter, "The Present Value ^ Relation: Tests Based on Implied Variance Bounds, Econometrica, XLIX (May, 1981), 555-574.
Leuthold, Raymond, "Random Walk and P r i c e Trends: The Live Cattle Futures Market," Journal of Finance, XXVII (September, 1972), 879-889.
Michener, Ronald, "Variance Bounds in a Simple Model of Asset Pricing," Journal of Political Economy, XC (February, 1982), 166-175.
Morgan, George, "Forward and Futures Pricing on Treasury Bills," Journal of Banking and Finance, V (December, 1981), 483-496.
Pesando, James, "On Expectations Term Premiums and the Volatility of Long-Term Interest Rates, Journal o Monetary Economics, XII (September, 1983), 467-474.
Poole, William, "Using T-Bill Futures to Gauge Interest-Rate Expectations," Federal Reserve Bank of San Francisco Economics Review, Spring, 1978, pp. 7 lb.
Puglisi, Donald, "Is the Futures Market for Treasury Bills Efficient?," Journal of Portfolio Management, IV (Winter, 1978), 64-67.
Rendleman, Richard and Christopher Carabini, "The Efficiency of the Treasury Bill Futures Market, Journal o— Finance, XXXIX (September, 1979), 895-914.
Richard, Scott and M. Sundaresan, "A Continuous Time _ Equilibrium Model of Forward Prices and Futures Prices in a Multigood Economy," Journal of Financial Economics, IX (December, 1981), 347-372.
Shiller, Robert, "The Volatility of Long-Term Interest ^ Rates and Expectations Model of the Term Structure, Journal of Political Economy, LXXXVII (December, 1979), 1190-1219.
97
Shiller, Robert, "The Use of Volatility Measures in Assess-ing Market Efficiency," Journal of Finance, XXXVI (May, 1981), 291-304.
, "Do Stock Prices Move Too Much to be Justi-fied by Subsequent Changes in Dividends?," American Economic Review, LXXI (June, 1981), 421-436.
Singleton, Kenneth, "Expectations Models of the Term Structure and Implied Variance Bounds," Journal of Political Economy, LXXXVIII (December, 1980), 1159-1176.
Tomek, William and Scott Querin, "Random Processes in Prices and Technical Analysis," Journal of Futures Markets, IV (Spring, 1984), 15-23.
Vignola, Anthony and Charles Dale, "Is the Futures Market for Treasury Bills Efficient?," Journal of Portfolio Management, V (Winter, 1979), 78-81.
Unpublished Materials
Huang, Roger, "Efficiency and the Volatility of Flexible Exchange Rates," unpublished manuscript, University of Pennsylvania, 1980.
Brochures
Inside T-Bill Futures, Chicago, Chicago Merchantile Exchange, 1983
The International Monetary Market, Chicago, Chicago Mercantile Exchange, 1982.
Treasury Bill Futures, Chicago, Chicago Mercantile Exchange, 1977.