leaders and followers among security analysts

Post on 04-Jan-2016

32 Views

Category:

Documents

1 Downloads

Preview:

Click to see full reader

DESCRIPTION

Graduate Project. Prepared by Li Wang Dept. of Statistics at McMaster University. Leaders and Followers among Security Analysts. Supervised by Dr. Veall and Dr. Kanagaretnam. Outline. Background and Data Description Performance of Security Analysts - PowerPoint PPT Presentation

TRANSCRIPT

Leaders and Followers

among Security Analysts

Prepared by Li WangDept. of Statistics at McMaster University

Supervised by Dr. Veall and Dr. Kanagaretnam

2

Outline

Background and Data Description Performance of Security Analysts Logistic Regression Analysis on Security Analysts Dataset Timeliness Analysts and Stock Price Conclusion and Discussion

3

Background on Security Analysts

Analyst

Brokerage Firm Institutional

clients

Individual Investor

Institution Investor

Wall Street Journal

Third Party

Buy, sell or hold

rank

rank

Background……

4

Continue Background…….

Firm A

Forecast 2

Forecast 4

Forecast 3

Forecast 5

Forecast 6

Forecast 7

Forecast 9

Forecast 11

Forecast 8

Forecast 10

Analyst 1

Analyst 3

Analyst 2

Analyst 1

Analyst 4

Firm B

Forecast 1

Leader

FollowerFollower

5

Institution Brokers Estimate System (I/B/E/S)

Adjustment file

Identifier File

Exchange Rate

Stopped Estimate

Report Currency

Excluded Estimates

S/I/G CodesBroker

Translations

Detail File

Actual File

Ticker

Data Description and Background……

6

Timeliness Leaders (Cooper, Day and Lewis, 2001)

Superior access to information Differential ability to process information Release forecast before competing analysts Herd behavior in finance market

Performance of Analysts

7

Leader and Follow Ratio (LFR)

F N N( , )2 2~

N : the number of earning forecastsToj : lead-timeT1j : follow-time

L F RT

To j

j

2

20

1 1

/

/

( Cooper et al, 2001)

T tj ii

N

0 01

T tj i

i

N

1 11

;

Hypothesis 1: Testing the forecast arrival times for leaders in pre-

release periods are greater than those in post-release periods.

Under null hypothesis θ0 = θ1, L F RT

To j

j

1

( Lawless, 1982 )

Performance of Analysts

8

A1

A2

A3

A4

A5

B5

B4

B3

B2

B1

-12 -10 -8 -7 -5 0 1 3 4 5 7 days

L F RT

Tj

j

0

1

1 2 1 0 8 7 5

1 3 4 5 7

5 2

2 02 6.

LFR > 1 The selected analyst is a leader.

LFR For Leaders

Forecast revision dates surrounding the forecast revision of a lead analyst

Performance of Analysts

9

LFR For Followers

Forecast revision dates surrounding the forecast revision of a follower analyst

A1

A2

A3

A4

A5

B5

B4

B3

B2

B1

-5 -4 -3 -2 -1 0 3 5 7 9 11 days

L F R

5 4 3 2 1

3 5 7 9 11

1 5

3 50 4 2 8 6.

LFR < 1 The selected analyst is a follower.

Performance of Analysts

10

Results of classification

Lead Analysts Percentage

Each firm over sample period (1994-2003) 13.68%

Each firm in a given year 28.10%

At least one firm leader in a given year 31.73%

Leaders in subsequent year 10.89%

11

Forecast Accuracy and Bias Percentage Forecast Error (Butler, Lang and Larry)

Forecast Bias: Signed Forecast Error

Standardize the Ranks to Scores (Hong et al, 2000)

Performance of Analysts

P F E F E A E A Eijt ijt j ( ) /

B F E A E F E A Eijt j ijt j ( ) /

ijtijt

ijt

R

N

1

1( Rijt = 1,…, Nit)

12

Hypothesis 2: the earning forecasts of followers are more accurate than those of leaders over the estimation year.

Performance of Analysts

Relative forecast accuracy of leaders and followers

Panel A : Half-year ahead earnings forecast errors and forecast bias:

Overall Leader follower T test M-W test

forecast accuracy:N

MeanMedian

7,1550.4930.500

1998 5157 0.486 0.512 -3.12*** -3.627***0.511 0.500

Bias:Mean

Median0.4940.500

0.490 0.494 0.52 -1.3350.500 0.500

13

Panel B: One- Quarter ahead of earnings forecast errors and forecast bias

Overall Leader follower T test M-W test

forecast accuracy:N

MeanMedian

3,9920.4740.500

1239 27530.470 0.476 0.65 -2.5670.500 0.500

% Bias:Mean

Median

0.485 0.500

0.471 0.491 1.87* -1.2350.500 0.500

Performance of Analysts

Relative forecast accuracy of leaders and followers

14

Boldness

Boldness: Absolute value of the difference between a particular forecast and the mean of outstanding consensus forecast.

■ Standardize to Scores: large deviation with higher scores and small deviation with lower scores.

Performance of Analysts

Forecast Issuance Timeline

Jun. Jul. Aug. Sept. Oct. Nov. Dec.

Pre-release period

Boldness

15

Hypothesis 3: A higher percentage of forecasts of leaders derivate from the consensus forecast compared to those of follower analysis.

Performance of Analysts

Table 2-6 Boldness of leaders and followers

Absolute Consensus Surprise

Overall Leader Follow T-test M-W test

N

Mean Median

7149 0.4954 0.5000

2021 5128 0.5225 0.4847 -5.44*** -5.758***0.5294 0.4808

16

Other AttributesPerformance of Analysts

Other Attributes of Leaders and Followers Overall Leader follower T test M-W test

Brokerage Size Mean Median

16.9414.00

17.47 16.75 -2.16** -2.468** 15.00 14.00

Forecast Frequency Mean Median

1.1320.882

1.119 1.136 1.03** -9.55* 0.879 0.882

Stock Coverage Mean Median

4.4082.000

2.393 2.447 1.16* -1.497* 2.000 2.000

Firm-Specific EXPR Mean Median

4.7154.000

4.79 4.68 1.2** -1.463*4.00 4.00

Business EXPRMean Median

8.0367.000

8.18 7.98 -1.45* -1.437* 7.00 7.00

17

Variable Specification of Security Analyst Dataset

Dependent Variable: Leader=1; Follower=0 Explanatory Variable: Forecast Accuracy (ACCUSCORE) Forecast Bias (BIASSCORE) Forecast Boldness (BDSCORE) Larger Brokerage Firm (LARGEBRKR ) Small Brokerage Firm (SMALLBRKR) Stock Coverage (COVER) Relative Forecast Frequency (RFREQ) Experience (EXPR) Following Analyst (NUMANALYST)

Model Fitting……

18

Summary Statistics of Analysts DatasetModel Fitting……

Variables

Mean Std. Deviation

Overall Leader Follower Overall Leader Follower

# of analysts 7341 1956 5385

BROKERSIZE 16.9384 17.4392 16.7415 12.9170 12.9315 12.9073

LARGEBRKR .2426 .2521 .2389 .4287 .4344 .4264

SMALLBRKR .2186 .1987 .2264 .4133 .3991 .4185

BUSYEAR 4.7149 4.7927 4.6842 3.6378 3.6503 3.6325

NUMANAYST 9.2213 9.0156 9.3021 5.3753 5.2158 5.4351

ACCUSCORE .4933 .5106 .4865 .3147 .2764 .3284

BIASSCORE .4935 .4903 .4947 .3184 .2824 .3315

RFREQ 1.1311 1.1191 1.1358 .6098 .5980 .6143

BDSCORE .4963 .5244 .4852 .2658 .2626 .2663

EXPR 1.7691 1.7927 1.7598 .6358 .6360 .6355

LFR 1.3037 2.2749 .9220 1.0993 1.4295 .6033

COVER 2.4079 2.3444 2.4329 1.7531 1.6875 1.7778

19

Logistic regression model has the form

Linear predictor

Modeling the conditional probability

Logistic Regression Model

Logistic Regression Analysis……

p Y y xx

x x

y y

( | )ex p [ ( )]

ex p [ ( )] ex p [ ( )]

1

1

1

1

( , )y 0 1

lo g( )

( )( ).

p x

p xX

1

( ) ,x x xd d 0 1 1

20

Logistic Regression Model for Analysts Dataset

Logistic Regression Analysis……

Coefficient Correlation

  䦋

Largebroker

Smallbroker cover

Numanalyst rfreq accu bias bold expr

largebroker 1 -.299(**) -.119(**) -.052(**) -.035(**) .006 .017 -.007 -.073(**)

smallbroker -.299(**) 1 -.040(**) .013 .032(**) -.007 .011 -.023 .032(**)

cover -.098(**) -.071(**) 1 .168(**) .034(**) .011 .008 .020 .209(**)

numanalyst -.045(**) .004 .161(**) 1 -.103(**) .028(*) .027(*) .022 .137(**)

Rfreq -.023(*) .017 .027(*) -.095(**) 1 .018 .003 .000 .049(**)

accuscore -.006 .004 .007 .016 .026(*) 1 .433(**) .052(**) .010

biasscore .010 .024(*) -.004 .013 -.001 -.020 1 -.012 -.002

bdscore -.006 -.025(*) .016 .020 .007 .076(**) -.033(**) 1 .034(**)

expr -.072(**) .032(**) .195(**) .111(**) .038(**) .006 -.014 .033(**) 1

• Pearson coefficients are above the diagonal line and Spearman coefficients are below the line. •***significant at 1% level, ** significant level at 5% level, * significant level at 10% level.

21

Analysis of Maximum Likelihood Estimation

Logistic Regression Analysis……

Parameter Estimate StandardError

WalkChiSq

Pr >ChiSq

OR 95% WaldConfid limits

INTECEPT -1.2182 0.1291 89.0212 <.0001

COVER -0.0372 0.0160 5.4334 0.0198 0.963 0.934 0.994

FREQ -0.0645 0.0414 2.4272 0.1192 0.938 0.864 1.017

EXPR 0.1227 0.0423 8.4142 0.0037 1.131 1.041 1.228

BOLDNESS 0.5410 0.0983 30.3107 <.0001 1.718 1.417 2.083

NUMANALYST -0.0119 0.00501 5.6637 0.0173 0.988 0.978 0.998

ACCU 0.3144 0.0995 9.9859 0.0016 1.369 1.127 1.664

BIAS -0.1303 0.0977 1.7792 0.1822 0.878 0.725 1.063

SMALLBRKR -0.1653 0.0677 5.9708 0.0145 0.848 0.742 0.968

LARGEBRKR 0.0218 0.0633 0.1188 0.0730 1.022 0.903 1.157

Goodness-of-fit: Hosmer and Lemeshow χ2=14.137 on 8 d.f., P=0.0483

22

Test the Hypothesis: Timeliness leader will have a higher impact on the stock price than followers.

Abnormal Excess Return: security return - value weighted

industry index in I/B/E/S. (EVENTUS)

Forecast Surprise: Forecast Revision:

Predecessor-based Surprise:

Consensus-based Surprise:

Timeliness and Stock Price……

F SC F E P F E

P F Eit

it i t

i t

( )

( )

1

1

F SC F E F E

F Eitit i

i

1

1

F SC F E C F

C Fitit t

t

( );

w here C FF E

ntit

i n

23

Analyst Timeliness and Contemporaneous Stock price

Hypothesis 4: The coefficient of regression of excess return (in short window) on forecast surprise for leader analysts is greater than the coefficient for follower analysts.

Where is the cumulative excess return over the two-day released period.

Timeliness and Stock Price……

E X R L eader F S F ollow er F Sijt ijt ijt ijt 1 2 3* *

E X R ijt

24

2-day Release Period Forecast Surprise Coefficients for Lead

and Follower Analysts

Timeliness and Stock Price……

Forecast Revision

Predecessor-based surprise

Consensus-based surprise

Coefficient(1)

F-value(2)

Coefficient(3)

F-value(4)

Coefficient(5)

F-value(6)

Intercept 0.0137 <0.0001 0.00038 <0.0001 -0.0046 <0.0001

Leader* FEijt0.0345 0.0345 0.135 <0.0001 0.382 <0.0001

Follow* FEijt-0.0182 0.0267 0.078 0.012 0.227 0.004

Adj. R2

0.03 0.056 0.253

N 7,025 7,350 6,987

25

Analyst Timeliness and Past Stock price

Hypothesis 5: Forecast surprises by leader analysts are not significantly correlated with the stock price performance during the period preceding the forecast revisions ( in long window). However, forecast surprises by follower analysts are positively correlated with excess return during the pre-revision period ( in long window).

Timeliness and Stock Price……

E X R L eader F S F ollow er F Sijt ijt ijt ijt 1 2 3* *

Where is the cumulative excess return over the 20 days pre- released period.

E X R ijt

26

20-Day Pre-Released Period Forecast Surprise Coefficients

for Leader and Follower

Timeliness and Stock Price……

Forecast RevisionPredecessor-

based surpriseConsensus-

based surpriseCoefficient

(1)P-value

(2)Coefficient

(3)F-value

(4)Coefficient

(5)T-value

(6)

Intercept 0.00034 <0.0001 0.0053 <0.0001 0.0274 <0.0001

Leader* FEijt0.0043 0.0023 0.023 0.134 0.0045 0.345

Follow* FEijt0.0012 0.0215 0.006 0.234 0.0034 0.0013

Adj. R2 0.083 0.067 0.087

N 6,945 6,752 6,894

27

Discussion and Conclusion:

There are quality differentials among security analysts in terms of the timeliness of their forecasts.

Lead analysts tend to be employed by larger brokerage firm and follow up fewer stocks than follow analysts.

Leaders are bolder than followers. Followers release more accurate forecasts than leaders since

leaders have to sacrifice the accuracy to be the first movers. Lead analysts identified by timeliness have a greater impact on

stock price than follower analysts. Forecast surprises by leader analysts are not significantly

correlated with the stock price performance during the period preceding the forecast revisions. However, forecast surprises by follower analysts are positively correlated with excess return during the pre-revision period.

Discussion and Conclusion……

28

Reference:

Rick A. Cooper, Theodore E. Day, Craig M. Lewis, 2000. Following the Leader: a study of individual analysts’ earnings forecasts. Journal of Finance Economics, 61, 383-416.

Brown, Lawrence D., 2001, How import is past analyst forecast accuracy?. Financial Analysts Journal 57,4-49

Brennan, M., and Subrahmanyam, A. 1995. Investment analysis and Price formation in securities markets. Journal of Financial Economics 38, 361-381.

Gleason, Cristi A., and Charles M. C. Lee, 2003, Analyst forecast revisions and market price discovery. The Accounting Review 78, 227-250.

Leone, A., and Wu, J., 2002. What does it take to become a superstar? Evidence from institutional investor rankings of financial analysts. Working paper, University of Rochester.

Milkhail, M., R. Willis and B. Walther, 1997. Do Security Analysts Improve Their Performance with Experience? Journal Accounting Research 35: 131-157.

Reference…..

29

Thank You!

top related