time delays, competitive interdependence and firm performance · 3 mcintire school of commerce,...

Strategic Management JournalStrat. Mgmt. J., 38: 506–525 (2017)

Published online EarlyView 20 April 2016 in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/smj.2512Received 4 November 2013; Final revision received 20 January 2016

TIME DELAYS, COMPETITIVE INTERDEPENDENCE,AND FIRM PERFORMANCE

JUKKA LUOMA,1* SAMPSA RUUTU,2 ADELAIDE WILCOX KING,3 andHENRIKKI TIKKANEN1,4

1 Department of Marketing, Aalto University School of Business, Helsinki, Finland2 Business Ecosystems, Value Chains, Foresight, VTT Technical Research Centre,Espoo, Finland3 McIntire School of Commerce, University of Virginia, Charlottesville, Virginia,U.S.A.4 Stockholm Business School, Stockholm University, Stockholm, Sweden

Research summary: Competitors’ experiences of prior interactions shape patterns of rivalry overtime. However, mechanisms that influence learning from competitive experience remain largelyunexamined. We develop a computational model of dyadic rivalry to examine how time delays incompetitors’ feedback influence their learning. Time delays are inevitable because the processof executing competitive moves takes time, and the market’s responses unfold gradually. Weanalyze how these lags impact learning and, subsequently, firms’ competitive behavior, industryprofits, and performance heterogeneity. In line with the extant learning literature, our findingsreveal that time delays hinder learning from experience. However, this counterintuitively increasesrivals’ profits by reducing their investments in costly head-to-head competition. Time delays alsoengender performance heterogeneity by causing rivals’ paths of competitive behavior to diverge.

Managerial summary: While competitive actions such as new product launches, geographicalexpansion, and marketing campaigns require up-front resource commitments, the potential liftin profits takes time to materialize. This time delay, combined with uncertainty surrounding theoutcomes of competitive actions, makes it difficult for managers to learn reliably from previousinvestment decisions. This results in systematic underinvestment in competitive actions. Theseverity of the underinvestment grows as the time delay between an investment and its positiveresults increases. Counterintuitively, however, competitors’ collective underinvestment increasesprofit-making opportunities. In industries with large time delays, companies that do invest incompetitive actions are likely to enjoy high returns on investment. It is also likely that rivals’paths of competitive behavior bifurcate. Together, these mechanisms generate large differences incompetitors’ profits. Copyright © 2016 John Wiley & Sons, Ltd.

INTRODUCTION

Competitive dynamics researchers often adopt abehavioral perspective (e.g., Chen et al., 2010a;Hsieh, Tsai, and Chen, 2014; Kilduff, Elfenbein,

Keywords: competitive dynamics; time delays; perfor-mance heterogeneity; behavioral strategy; reinforcementlearning*Correspondence to: Jukka Luoma, Department of Marketing,Aalto University School of Business, P.O. Box 21230, 00076Aalto, Finland. E-mail: [email protected]

Copyright © 2016 John Wiley & Sons, Ltd.

and Staw, 2010; Marcel, Barr, and Duhaime, 2010),recognizing that managers must make decisionsconcerning rivalry in the absence of a clear under-standing of how their firm’s competitive actionstranslate into performance outcomes. This lack ofclarity prohibits treating competition as a mere opti-mization problem and motivates managers to basetheir decisions on the outcomes of earlier choices(Cyert and March, 1992; Lamberg et al., 2009). Aslearning from prior experiences can influence man-agers’ decisions, identifying factors that affect the

Time Delays and Competitive Interdependence 507

process of learning from experience can help under-stand how competitive interaction patterns formover time (Chen and Miller, 2012). This can, in turn,facilitate addressing fundamental questions in strat-egy such as how and why performance differencesemerge among firms.

Building on this insight, we examine time delaysin interfirm rivalry as an antecedent of differencesin industry profitability and performance hetero-geneity among competitors. Time delays have inter-ested competitive dynamics scholars for the entiretyof the field’s existence (e.g., MacMillan, McCaf-fery, and Van Wijk, 1985), and such delays are anatural focus of inquiry for studying the factorsthat affect learning in rivalry. Scholars recognizethat time is needed to execute competitive actions(Chen and Hambrick, 1995; Miller and Chen, 1994)and to realize the results of these actions (Bridoux,Smith, and Grimm, 2013). Such time delays canaffect firm performance. For example, research hasdemonstrated that slow competitors are usually ata disadvantage in capturing contested opportunities(e.g., Boyd and Bresser, 2008; Chen and Hambrick,1995; Ferrier, 2001; Hawk, Pacheco-De-Almeida,and Yeung, 2013; Miller and Chen, 1994). However,to date, research has focused on the immediate eco-nomic consequences of time delays. The effects oftime delays on longitudinal patterns of competitiveinteraction have not been systematically studied.

We employ computational modeling to uncoverhow time delays shape rivalry beyond the short termby influencing managers’ learning from previouscompetitive interactions. Our findings are consistentwith the existing literature on learning in showingthat time delays hinder learning from experience(Rahmandad, Repenning, and Sterman, 2009; Ster-man et al., 2007). In particular, time delays reducethe perceived reward of investing in competitiveactions, causing companies to underuse them. How-ever, this effect may counterintuitively enhancecompetitors’ performance because, at a collectivelevel, time delays cause firms to invest less incostly head-to-head competition. The positive per-formance effect of time delays has gone unnoticedto date because prior research on time delays inlearning has not considered the impact of com-petitors’ decisions on a focal firm’s performance(e.g., Rahmandad, 2008; cf., Katila and Chen,2008). We also show how the impact of time delayson learning fosters performance heterogeneity bycausing rivals’ behavioral trajectories to diverge.Overall, our study establishes time delays firmly

alongside other explanations of industry profits andperformance heterogeneity (e.g., Lenox, Rockart,and Lewin, 2010). The paper also advances ourunderstanding of how micro-level behavioral mech-anisms shape macro-level patterns of competitiveinteraction over time (e.g., Chen and Miller, 2012).

THEORETICAL BACKGROUND

Our study investigates the phenomenon of interfirmrivalry, which is a fundamental issue in strategicmanagement. In line with the literature on compet-itive dynamics (Chen and Miller, 2012: 137), weconceptualize interfirm rivalry as a series of com-petitive actions and interactions. In this study, com-petitive actions are defined as product market movesthat aim to increase the attractiveness of a firm’sofferings in the eyes of customers relative to com-petitors’ products and services (Miller and Chen,1994). Such actions might include product intro-ductions that provide new benefits to customers,geographical expansion initiatives that bring a firmcloser to customers, or marketing campaigns thatbuild a favorable brand image.

The competitive moves that rivals make todayhinge on their history of previous competitiveencounters (e.g., Kilduff et al., 2010), as firms’previous competitive experiences of past successshape their current decision making (Lamberg et al.,2009). Any factor that influences firms’ learningfrom experience may have indirect but cumulativelysignificant implications for how and what compet-itive interaction patterns emerge. We propose thattime delays are one such factor with the potential toshape rivalry over repeated competitive interactionsby affecting managers’ learning from experience.

The strategic significance of time is well estab-lished in the competitive dynamics literature(Ferrier, 2001). Scholars have investigated, forexample, how lags in competitive responses (Boydand Bresser, 2008), in the speed of decision making(Eisenhardt, 1989), and in the temporal spacingof actions (Laamanen and Keil, 2008), as well asinertia in “altering [a] competitive stand” (Millerand Chen, 1994: 2), affect the performance ofcompeting firms. Scholars also recognize that thefull performance consequences of competitiveactions do not materialize immediately and thatthere can be differences in the speed at which dif-ferent competitive moves affect firm performance(Bridoux et al., 2013).

Copyright © 2016 John Wiley & Sons, Ltd. Strat. Mgmt. J., 38: 506–525 (2017)DOI: 10.1002/smj

508 J. Luoma et al.

We consider the role of time in interfirm rivalry(Ancona et al., 2001) from the perspective of learn-ing. In particular, we examine how rivalry-relatedtime delays influence competition by delaying thefeedback that a firm receives about its decisionsconcerning competitive actions. In general, the sig-nificance of delayed feedback from the perspec-tive of learning is well established. For example,the cognitive “myopia” of managers can constrainlearning from temporally distant outcomes (Diehland Sterman, 1995; Levinthal and March, 1993).Moreover, stakeholders often pressure executives totake actions based on short-term outcomes, whichmay be counterproductive if success presupposespatience (Aspara et al., 2014). Organizational learn-ing may also be compromised when time delaysbetween decisions and outcomes span beyond thetenures of key decision makers (Paich and Sterman,1993).

Our analysis focuses on two rivalry-related timedelays. First, market reaction lag refers to the timedistance between the launch of a competitive actionand observable changes in customer behavior inresponse to shifts in the relative attractiveness ofrivals’ offerings. A related term in marketing is“wear-in time” or “the lag before the peak impacton sales is reached” following a marketing action(Srinivasan, Vanhuele, and Pauwels, 2010: 680).Market reaction lags are positively associated withthe “time to positive performance impact” (Bridouxet al., 2013: 931). Various factors can contribute tomarket reaction lags. For example, customers donot immediately change their purchasing behaviorwhen new products and services become availablebecause they may be either unaware of the newproduct or service or uncertain of its benefits(Horsky, 1990). This is the case especially ifproduct and price comparisons are difficult or ifswitching costs are high (Chatain and Zemsky,2011). Low decision-making frequency might alsocontribute to delayed sales responses to compet-itive actions; for example, potential customerswho reconsider their service subscriptions once ayear will respond to a change in price levels moreslowly than customers who revisit their decisionsbimonthly. Finally, customers may choose to delayresponses to new products or price cuts in the hopeof seeing even better products launched and furtherprice cuts (Horsky, 1990). Because these lagslargely reflect buyer characteristics and behaviors,market reaction lags are generally industry-wide;that is, competitors experience similar delays

in how quickly the market responds to theiractions.

Second, we consider execution delays thatstem from the process of carrying out competitiveactions. Several barriers to competitive actionsexist that contribute to execution delays. Forexample, managers and specialists often requiretime to amass various customer, competitor, andtechnology-related knowledge to help determinethe specifics of their competitive actions (e.g.,product design, advertising media) (cf., Ferrier,2001). Moreover, the launch of complex productsinvolves time-consuming coordination acrossdepartments (MacMillan et al., 1985). Finally,the execution of competitive actions may requiredealing with public authorities or renegotiatingwith suppliers, which further delays the launchof competitive actions. Firms possess differentiallevels of ability to address these barriers to action(e.g., Hambrick, Cho, and Chen, 1996; Hawket al., 2013). For example, some firms may havebetter information-processing systems that helpthem more quickly resolve uncertainty (Eisenhardt,1989); some companies have developed organiza-tional routines that facilitate intra-organizationalcoordination (Becker, 2004); and some organi-zations have more established relationships withexternal parties, reducing inter-organizationalnegotiation and coordination. This leads us toassume that execution delays are generally firmspecific.

The existing competitive dynamics literature hasnot extensively discussed the performance implica-tions of market reaction lags (see, however, Bridouxet al., 2013). But when regressing firm performancevariables on lagged product market actions, schol-ars assume that market reaction lags exist (e.g.,Young, Smith, and Grimm, 1996: 250). With regardto execution delays, the existing literature consid-ers them typically problematic. In intensive rivalrywhere opportunities are contested and thereforeshort-lived (Chen et al., 2010a; Hambrick et al.,1996; Hawk et al., 2013; Smith et al., 1991), exe-cution delays increase the risk of preemption byfaster competitors (Hawk et al., 2013). Further-more, the sluggish execution of competitive actionscan hamper the release of committed organizationalresources to other uses, thus reducing strategicflexibility.

Although existing research acknowledges theexistence of time delays in rivalry and recognizestheir potential impact on rivals’ performance (e.g.,



Boyd and Bresser, 2008), scholars to date havefocused on the immediate or near-term economicconsequences of time delays. To extend this lineof inquiry, we employ computational modeling—inparticular, system dynamics (Sterman, 2001)—touncover how time delays shape patterns of rivalryand associated performance outcomes over time byaffecting firms’ learning from previous competitiveinteractions.

Our model builds upon the literature on rein-forcement learning in psychology and management(e.g., Atkins, Wood, and Rutgers, 2002; Diehl andSterman, 1995; Lam et al., 2011; Lurie and Swami-nathan, 2009; Rahmandad, 2008; Rahmandad et al.,2009), which emphasizes actors’ tendency to movein the action space in directions that increase theperceived reward. Causal ambiguity impedes butdoes not completely inhibit managers’ ability toattribute changes in performance to their decisions(Lippman and Rumelt, 1982; Mosakowski, 1997).Building on research concerning the psychology oflearning (Shanks, Pearson, and Dickinson, 1989;Wasserman and Miller, 1997), we assume that man-agers use the temporal proximity of their decisionsto changes in performance in discerning which deci-sions are responsible for particular performanceoutcomes.

The influence of time delays in performancefeedback on the lessons that managers derive fromtheir previous decisions has received the attentionof numerous management scholars over the pastfew decades (King, 2007; Moxnes, 1998; Rahman-dad, 2008; Rahmandad et al., 2009; Sterman, 1989;Sterman et al., 2007). For example, resource-basedscholars have argued that the temporal distancebetween an investment in capabilities and the asso-ciated performance outcomes hinders firms’ abilityto discern which capabilities underlie the successof high-performing firms. Whereas experience gen-erally reduces causal ambiguity about the driversof performance (Mosakowski, 1997), time delayscan limit the utility of learning from experience.King (2007: 170) proposes that time delays havethe potential to foster performance heterogeneity byobstructing managers’ capacity to understand whatdrives the performance of their rivals or of their ownfirms.

The existing management literature has focusedon explaining how time delays in decision-makingfeedback produce various decision-making patholo-gies. Empirical and computational evidence sug-gests that time delays obstruct managers’ ability to

connect strategic decisions with performance out-comes, potentially leading to behavioral oscillationand suboptimal courses of action (Denrell, Fang,and Levinthal, 2004; Rahmandad, 2008; Rahman-dad et al., 2009). In normative terms, these resultssuggest that managers seeking to enhance their abil-ity to learn from feedback should minimize the tem-poral distance between their decisions and observedperformance outcomes.1 This could be achieved, forexample, by creating appropriate information sys-tems (Eisenhardt, 1989) or by designing organiza-tional structures that ensure rapid information flows(Smith et al., 1991).

We build on these insights to explore howrivalry-specific time delays shape competitive inter-action patterns by making it more difficult for man-agers to identify productive courses of competitivebehavior based on performance feedback. We thusextend the competitive dynamics literature, whichhas not systematically examined the learning impli-cations of time delays in rivalry, as well as thelearning literature, which has not considered theimplications of delayed feedback in competitivesettings.

MODEL DEVELOPMENT

Conceptual overview

Our model (see Figure 1) focuses on dyadicrivalry, which unfolds over time as a series ofinter-temporally connected decisions about com-petitive actions (Ferrier, 2001). Each decisionrequires a rival to anticipate trade-offs between theexpected expenses and sales associated with per-forming competitive actions. Competitive actions,which are costly to execute, may generate salesby tapping into poorly catered market needs, bytargeting current customers to prevent them fromswitching to competitors’ offerings or by targetingrivals’ customers. However, managers cannotreliably estimate the performance outcomes of theircompetitive actions ex ante. Competitive actionsintroduce something novel to the market, whichalways triggers an element of unpredictability interms of both customers’ responses and the actual

1 This generalization is not entirely without boundary conditions.Noisy feedback (Lurie and Swaminathan, 2009) and cognitivecapacity constraints (Lam et al., 2011) limit the benefits offrequent performance feedback.


510 J. Luoma et al.

costs required to carry out the action. Thus, theuncertainty surrounding the performance outcomesof a firm’s competitive actions is Knightian ratherthan probabilistic in nature (see, e.g., Mosakowski,1997: 438).

Consequently, we assume that competitors’learning processes are backward looking, asrivals cannot know “the value of a strategy notyet chosen” (Greve, 2002: 5). Feedback shapesdecisions about the future through the mechanismof reinforcement learning. The temporal proximitybetween a decision and an observable changein performance is used as a cue for causality(Shanks et al., 1989; Sloman, 1996; Wassermanand Miller, 1997). In each decision-making event,which occurs at specific time intervals (Gersick,1994), a firm evaluates its previous decision andthen adjusts its competitive behavior until the nextdecision-making opportunity occurs. The firm’sdecision variable is the number of competitiveactions to be undertaken over a given time period.This is referred to as firm-level competitive activity.For a dyad over a given time period, the intensityof dyad-level rivalry refers to the sum of firm-levelcompetitive activity of both rivals (Young et al.,1996).

The effects of competitive actionson performance

We model competition as an ongoing struggleto maintain attractiveness in the eyes of cus-tomers rather than as a one-shot positioning game(Hotelling, 1929). Unless a firm continues to investin costly competitive moves, its attractivenessin the eyes of customers, relative to competitivealternatives, will erode. We build an attractionmodel to capture these dynamics. This family ofmodels has long been used to model competingfirms’ market shares (Farris et al., 1998; Nakanishiand Cooper, 1974). In our model, customers arewilling to pay for a firm’s products or services tothe extent that the firm takes actions that maintainor increase the attractiveness of those offerings.The firm’s competitive behavior is modeled as acontinuous, potentially varying stream of compet-itive actions, xi. Customer choice is modeled as aprobability that is a function of a firm’s decisionsto take competitive actions. The more competitiveactions a firm takes, the higher the likelihood thatcustomers are attracted to that firm instead of to itscompetitors. Specifically,

Pi =(xi∕x0

)𝛼 ∕[Σj

(xj∕x0

)𝛼 +(x𝜙∕x0

)𝛼](1)

Parameter x𝜙 reflects the attractiveness of thenone-of-the-above alternative, including the optionof saving the money for later use (Lenox, Rockart,and Lewin, 2006), and x0 is the reference value ofattractiveness. Customers’ sensitivity to the com-petitive actions of firms is reflected in 𝛼. Equation 1implies that customers generally prefer to trans-act with firms that take competitive actions toincrease their attractiveness relative to competitors.However, for reasons detailed earlier, customers’responses to changes in the attractiveness of avail-able products and service are delayed. To opera-tionalize these market reaction lags, denoted by 𝜏m,we define the firm’s market share, MSi, using a dif-ferential equation:

dMSi∕dt =(Pi − MSi

)∕𝜏m (2)

Equation 2 states that a firm’s market shareapproaches a level determined by its level ofcompetitive activity, but this occurs gradually ascustomers become aware of the attractiveness ofthe firm’s offering and accordingly change theirbehavior.

Our model assumes that a firm must makeup-front investments prior to undertaking any com-petitive actions. Changes in the firm’s level ofinvestment take time to affect the level of compet-itive activity realized; this execution delay of thefirm is denoted by 𝜏e. For example, although a deci-sion to launch new products will quickly increasea firm’s product development costs, the new prod-ucts themselves will not be immediately availableto customers. The execution delay reflects the timebetween when the decision is made and when theproduct is launched. Likewise, if a firm decidesto reduce its investment in competitive actions, itsup-front spending will decrease, but its competitiveactivity will not decline immediately; because pre-vious investments in ongoing projects have alreadybeen committed, those projects will be completed.To capture these assumptions, the cost of competingdepends on the desired (or intended) level of com-petitive activity, xi

*. The cost of competitive actions,Ci, is assumed to be constant. Execution delay ismodeled using a differential equation. The firm’slevel of competitive activity, xi, approaches a leveldesired by management, xi

*, where 𝜏e determinesthe speed of approach. In sum, the level of compet-itive activity and the consequent profit of firm i, 𝜋i,



DECISION MAKING

• Increase or decrease competitive activitty

Execution delay

COMPETITIVE ACTIONS

• Level of competitive activity

Market reaction lag

FIRM PERFORMANCE

• Profit

COMPETITORS’ ACTIONS

• Level of competitive activity

Figure 1. The firm’s decision-making process

are given by the following:

dxi∕dt =(x∗i − xi

)∕𝜏e (3)

𝜋i = S · MSi –Ci · x∗i (4)

where the largest possible market size is set at S.An essential assumption in our model is that the

information in Equations 1–4 is not transparent tomanagers. Instead, consistent with the notions ofselective attention and causal ambiguity, managershave an imperfect understanding of what they havedecided previously and how those decisions havebeen reflected in changes in performance. Thislimited understanding forms the basis for futuredecisions about competitive actions.

The decision process

The decision process that determines firm-levelcompetitive activity is triggered at predefined timeintervals, T . Because decisions take time to affectperformance, it is better for a firm to evaluate itsdecisions after some time has passed rather thanto evaluate them immediately. During each roundof decision making, a firm performs a “search” tochange its competitive behavior. Managers’ deci-sions pertain to the direction of change in theirfirm’s level of competitive activity. The firm maybecome competitively more active by exploringnew opportunities or by capturing market sharefrom rivals (Di(t)= 1). Alternatively, the firm mayreduce the use of competitive actions to lower costs(Di(t)= -1). If a firm increases its activity and thefirm’s sales revenues increase more than the costsof increased competitive activity, then competitiveactivity is more likely to increase in the future. Sim-ilarly, if the firm decreases competitive activity and

the cost savings exceed the lost revenues, then fur-ther reductions in competitive activity become morelikely.

More formally, the propensity of a firm toincrease (decrease) its competitive activity,P[Di(t)= 1] (=1 – P[Di(t)= -1]) and the desiredlevel of competitive activity are updated whent=T , t= 2 T, etc., in the following manner:2

P[Di

S (t) = 1]= w · P

[Di (t − T) = 1

]

+ (1 − w) ·(FBi + 1

)∕2 (5a)

x∗i (t + h) = x∗i (t + h − T) + Li (t) · Di (t) (5b)

The magnitude of the change in competitiveactivity, Li(t), is drawn from a Poisson distributionwith a mean of one. The stochastic component in thedetermination of the search step length accounts foridiosyncratic firm-specific situational factors thatinfluence the scope of the search.3 The shape ofthe Poisson distribution captures our assumptionthat small changes in competitive activity are morelikely to occur than large changes. The extent of

2 The precise sequence of events is that the firm first evaluatesits previous decision (Equation 5a) and then changes its behaviorin light of this information (Equation 5b). The time differencebetween these events is very small, h. (Simultaneous updatingof beliefs P[Di(t) = 1] (Equation 5a) and behavior xi*(t) wouldcause a simultaneity problem.) We let h= 0.5, which is also thetime step of our numerical integration algorithm.3 A fixed step size, the alternative assumption, would require avery strong assumption that firms’ search behavior is entirelydetermined by external factors. The assumption of a fixed step sizealso leads to unrealistic model behavior. For example, when w iszero, a fixed step size induces firms to change their competitivebehavior in perfect synchrony, which generates a “cooperationsolution”; that is, an intensity of competition that maximizesthe dyad’s aggregate profits, which is an unlikely and unstableoutcome in practice.


512 J. Luoma et al.

a firm’s reliance on historical vs. current feedbackin determining the firm’s search direction is cap-tured by a weight parameter, w. We tested multi-ple values of the parameter between zero and one.These experiments did not alter our theoretical con-clusions about how time delays influence learning.Thus, we report the results obtained with a represen-tative value (w= 0.75). Feedback on the appropriatesearch direction, FB i, is defined as follows:

FBi=sign{[𝜋i (t) −𝜋i (t − T)

]·[x∗i (t) − x∗i (t − T)

]}

(6)The function sign {·} equals one (minus one) if

the expression is greater (less) than zero. If the levelof competitive activity or profit has not changed, themodel assumes that a firm’s propensity to increaseits competitive activity will approach indifference(i.e., FBi = 0).

Equilibrium analysis

We first calculate an analytical solution as a bench-mark for our simulation analyses by assumingtwo rational profit-maximizing firms. Our assump-tions about rationality are familiar to (neoclassical)economists: a firm perfectly understands demandand customer behavior, knows the costs of its owncompetitive actions and those of its rival firm, aimsto maximize its profits, and assumes that its rivaldoes the same. For simplicity, we also assume thatthere are no delays in the system, and we assume,as in the simulations that follow, that 𝛼 = 1. Underthese assumptions, the profit of firm i is simply thefollowing:

𝜋i = S · xi∕(Σjxj + x𝜙

)− Ci · xi (7)

The Nash equilibrium level of competitiveactivity is derived by finding the xi in a systemof equations where the partial derivatives ofEquation 7 are set to zero, that is, 𝜕𝜋i/𝜕xi = 0for i= {1,2}. We use Mathematica to perform thesymbolic computation. The system of equations hastwo solutions, one of which is positive (competitiveactivity cannot be negative). Thus, the resultingequilibrium level of competitive activity for firm iis the following:

xNEi =

{−2C2

i · x𝜙 + Cj ·√

S ·√[(

4x𝜙·(Ci + Cj

)+ S)] − 2Ci · Cj · x𝜙 + Cj · S

}

∕2(Ci + Cj

)2(8)

The equilibrium solution (Equation 8) is derivedwithout time delays, but it is also the (Nash) equi-librium solution for the system with time delays.Consider that firms optimize their levels of compet-itive activity with respect to the profits obtained in asteady state. Steady-state profits are not affected bytime delays (Equations 1–4). Hence, the optimallevel of competitive activity does not change withthe time delays in the model. We assume thatrational actors can perfectly predict how currentdecisions affect future performance; because fullinformation is already available, there is no learningover time. Thus, rational profit-maximizing firmswill always choose xi

NE (Equation 8), regardlessof the time delays. We ensure that a firm’s optimalpolicy does not change even when we manipu-late the time delays in the model by holding allparameters in Equation 8 equal across simulations.

Design of simulation experiments

Our experimental design ensures that profitabilitydifferences both across and within simulated indus-tries have purely “behavioral” origins. To achievethis goal, we assume that the industry is “objec-tively” the same in all simulations. The industryparameters are fixed across simulations; therefore,if competitors were rational, all simulations shouldproduce identical levels of competitive intensity andindustry profits. Similarly, we assume no resourceheterogeneity between rivals by designating equalaction costs across firms (i.e., Ci =Cj). In this situa-tion, two rational rivals would choose the same levelof competitive activity and obtain identical levels ofperformance.

While the firms’ optimal policy, xiNE, is the same

across simulations and firms, time lags affect thefeedback that managers receive, which influenceswhat managers perceive to be the reasonable courseof action. For example, managers do not know howquickly additional R&D inputs will increase theoutput of marketable products (execution delay),nor can they be certain how quickly customers willreact to an advertising campaign (market reactionlag). Prior literature suggests that firms have limitedknowledge of time delays and short-termism pre-vails (Aspara et al., 2014). To operationalize theseassumptions, the delay parameter ranges are setto ensure that the upper values of delays are longrelative to the decision-making frequency. Specifi-cally, firms follow a decision-making cycle of fivetime periods, and time delays are assumed to vary



between 0 and 10 periods. When feedback is imme-diate, the following modified equations apply:

Market reaction lag = 0 ⇒ MSi (t) = Pi (t) (2′)

Execution delay = 0 ⇒ xi (t) = x∗i (t) (3′)

Otherwise, the model is run according toEquations 1–6. To investigate whether the impli-cations of time delays depend on the fact thatwe model dyadic rivalry in particular and notmerely decision making in general, we compare theresults for the dyadic simulations with simulationresults obtained for an “independent firm.” Anindependent firm performs a search in a competi-tive environment in which the rival firm does notchange its behavior. Independent firm simulationsrepresent the firm-centric assumption, commonlymade in the behavioral strategy literature, thata focal firm’s strategy and performance can bemodeled in isolation from its rivals’ strategies (e.g.,Katila and Chen, 2008).4

The model’s payoff structure resembles a Pris-oner’s Dilemma. Specifically, the Nash equilibriumis calibrated at 24 actions per firm per time period,with S= 100, x𝜙 = 12, Ci = 1 and 𝛼 = 1. With thisparameter configuration, the dyad’s aggregatedprofits (Equations 1 and 4) would be maximizedif both firms reduced their levels of competitiveactivity to 11.3 actions per firm per time period.Beyond this point, further reductions in the inten-sity of rivalry do not improve competing firms’performance because the outside-dyad competitivepressure (x𝜙 > 0) causes rivals’ sales to decrease ifthey employ competitive actions very infrequently.The model is run for 1,000 time periods. The modelparameters and the values used in the reportedsimulations are summarized in Table 1. We measurethe firms’ behavior and performance at the endof each simulation run. Using end-of-simulationvalues (after a long simulation run) ensures thatany transient dynamics caused by the initialconditions do not affect the results. Moreover,end-of-simulation values capture the full range

4 This assumption may indeed be reasonable, for example, notonly in monopolies but also in fragmented or emerging industrieswhere the focal firm’s performance is not substantially affected byany single rival’s decision making. In contrast, dyadic simulationscorrespond to oligopolistic industries where the performance ofone firm significantly depends on the competitive actions of rivals.

of behavioral outcomes caused by the model,which would be suppressed if we calculated onlywithin-simulation averages or sums.5

RESULTS

In the base case, we assume that (1) the firm isnot affected by the decisions of a rival (indepen-dent firm) and that (2) there are no time delaysin performance feedback. Under these simplifiedconditions, the assumed behavioral rules govern-ing firms’ decision making lead to behavior thatresembles the behavior of an economically ratio-nal actor. We then relax the simplifying assumptionsby allowing both competitive interdependence andtime delays to influence learning from feedback.The ensuing simulations reveal systematic depar-tures from what we would expect from a ratio-nal, profit-maximizing firm. We first show how theassumption of competitive interdependence affectsfirms’ learning from feedback. Second, we examinehow time delays affect the learning of an indepen-dent firm. Finally, we consider the interactive effectsof these two mechanisms.

Learning dynamics

Competitive interdependence and learning

Figure 2 illustrates that competitive interdepen-dence obstructs learning from feedback, whichleads to performance heterogeneity. Two simula-tions are shown. A baseline model is first run withthe independent firm assumption. This is opera-tionalized by assuming the rival firm’s competi-tive activity to be a constant at 24 actions per timeperiod. When no rival obstructs the focal firm’slearning from feedback, the focal firm’s level ofactivity displays modest oscillation surrounding theoptimal policy (i.e., 24 actions per time period),and the oscillation of performance is nearly indis-tinguishable from the performance obtained fromthe equilibrium analysis above. In other words,the independent firm’s behavior and performanceare reasonable approximations of the behavior andperformance of a rational profit-maximizing firm.With this established, the model is then run withan assumption of dyadic rivalry. The rival firm

5 Vensim DSS for Windows Version 6.1c Double Precision wasused to run the model.


514 J. Luoma et al.

Table 1. Summary of model parameters

Parameter Default Description

Rewards and costs of competitive actionsx𝜙 12 Attractiveness of the none-of-the-above option (action/period)x0 10 Reference value of attractiveness (action/period)𝛼 1 Sensitivity of customers to the firm’s competitive actionsCi 1 Unit cost of competitive actions (€/action)S 100 Total market size (€/period)x∗i (0) 24 The firms’ initial levels of competitive activity (action/period)Decision making about competitive actionsw 0.75 Weight parameter that determines the extent of a firm’s reliance on historical vs.

current feedbackE[Li(t)] 1 Mean search step size (action/period)P[Di(0)= 1] 0.5 Propensity to increase competitive activity at the start of the simulationT 5 Decision-making time interval (period)Time delay𝜏m {0, 1, … , 10} Market reaction lag (period)𝜏e {0, 1, … , 10} Execution delay (period)Simulation controlFinal time 1,000 Length of the simulation (period)

is allowed to change its initial level of competi-tive activity (24) according to the same behavioralrules that govern the focal firm’s decision mak-ing. In contrast to the independent firm simulation,the interaction between rivals’ learning processesgenerates large and relatively enduring differencesin firm-level competitive activity between firms,which result in significant performance differences.

There are two mechanisms by which interdepen-dence between rivals affects their learning fromfeedback. First, a rival firm’s level of competi-tive activity affects the focal firm’s payoff land-scape. The focal firm’s optimal level of activity firstincreases and then decreases with the rival firm’slevel of activity. Initially, as the rival firm increasesits competitive activity, the focal firm’s optimallevel of competitive activity increases. The intu-ition is that the focal firm must initially take morecompetitive actions to mitigate the erosion of prof-its as a result of lost sales revenues (Derfus et al.,2008). As the rival firm further increases its com-petitive activity, maintaining sales becomes increas-ingly costly, which then deters the focal firm frominvesting in competitive actions (Ferrier, Smith, andGrimm, 1999).

Second, concurrent changes in rivals’ levels ofcompetitive activity confound efforts to learn fromfeedback. During any given decision event, anincrease in a rival firm’s competitive activity exertsdownward pressure on the focal firm’s performance.

As a result, the focal firm is more likely to expe-rience a performance decline, regardless of thefocal firm’s decision. For example, if the focal firmdecides to increase activity, the advantages of itssales growth will be neutralized by the rival’s con-current increase in competitive activity and willtherefore be unlikely to offset the costs incurredby additional investments in competitive actions.Alternatively, a rival’s increased activity will exac-erbate the diminishing sales that result from afocal firm’s simultaneous decision to decrease activ-ity, thus making revenue losses more likely tooutstrip cost savings. Similar learning challengesarise when a rival firm decreases its competitiveactivity.

Time delays and learning

Time delays increase the difficulty of learning fromfeedback, leading to systematic departures from theoptimal level of investment in competitive actionsand to large differences between simulation runs.As shown in Figure 3, an independent firm’s behav-ior is relatively close to the optimal policy (i.e., theNash equilibrium) when feedback is prompt. How-ever, if decisions about competitive actions and theirobservable performance outcomes are not close toone another in time, then the firm deviates system-atically from the optimal policy. Specifically, timedelays cause an independent firm’s level of activityto be both (1) lower on average and (2) more varied



(a)

(b)Competitive activity

Profit

0

10

20

30

40

50

60

0 100 200 300 400 500 600 700 800 900 1000

Com

peti

tive

act

ivit

y (xi)

Independent firm Focal firm (dyadic rivalry)

Rival firm (dyadic rivalry) Nash equilibrium

0

5

10

15

20

25

30

0 100 200 300 400 500 600 700 800 900 1000

Pro

fit (�i)

Independent firm Focal firm (dyadic rivalry)

Rival firm (dyadic rivalry) Nash equilibrium

Figure 2. Independent firm vs. dyadic rivalry. (a) Competitive activity and (b) profit

over time. These patterns are consistent with earlierresearch and the intuition that time lags hinder theability of a firm’s managers to reliably make associ-ations between decisions about competitive actionsand performance outcomes, leading to oscillatingand suboptimal behavior (e.g., Denrell et al., 2004;Rahmandad et al., 2009). Figure 3 demonstrates theconsequences of time delays by showing the effectof selected time-delay configurations on an inde-pendent firm’s behavior.

Time delays affect learning from feedback intwo ways, both of which reduce an independentfirm’s average level of competitive activity. First,time delays can cause decision makers to makepremature, erroneous conclusions that investmentsin competitive actions are mistakes. Because weassume that the costs of competitive actions gener-ally materialize before the rewards do, a decisionto increase competitive activity results in a tran-sient decline in performance. Consequently, firms


516 J. Luoma et al.

0

5

10

15

20

25

30

35

0 100 200 300 400 500 600 700 800 900 1000

Com

peti

tive

act

ivty

(x i

)

Time

No time delay Market reaction lag = 3

Execution delay = 4 Nash equilibrium

Figure 3. Independent firm’s behavior with selected time-delay configurations

are more likely to reverse investment decisions thatwould in fact improve performance in the longterm. Second, time delays increase a firm’s ten-dency to repeat decisions to reduce spending oncompetitive actions. Reducing spending on com-petitive actions lowers costs immediately, whereasthe decline in sales is not observed until aftera delay. Consequently, with time delays, a firmis less likely to recognize that reduced spend-ing on competitive actions is causing it to losecompetitiveness.

In addition to the effects on a firm’s averagelevel of competitive activity, time delays lead tooscillation in competitive behavior because theeffects of previous decisions carry over to influencefeedback for subsequent decision-making rounds.This learning challenge has been called “temporalcomplexity” (Rahmandad, 2008). Because deci-sions to take action materialize after an executiondelay and because the performance effects ofactions are realized over multiple time periods,current levels of performance are the result ofdecisions made over multiple decision-makingrounds. Thus, time delays introduce the challengeof discerning the effect of earlier decisions fromthe performance effects of the decisions made morerecently, and this challenge increases the variationin firm-level competitive activity.

Time delays and performance

The competitive dynamics and learning streamsof literature emphasize the negative effect of timedelays on firm performance (e.g., Chen et al.,2010a; Ferrier, 2001; Rahmandad, 2008; Rahman-dad et al., 2009; Sterman et al., 2007). Althoughour model is capable of producing this result, themodel also reveals additional, previously unnotedinsights about how time delays influence firmperformance. In particular, time delays can reducethe intensity of industry competition and thus havea positive effect on firm profits.

In general, when a firm’s performance feed-back is delayed, it tends to underinvest in com-petitive actions relative to the available profitablemarket opportunities. The implications for perfor-mance, however, depend critically on our assump-tions regarding competition and time delays. Timedelays always reduce firm performance in the inde-pendent firm condition, which assumes that firmperformance is not affected by a competitor’s deci-sions. However, in contrast to established wis-dom about the detrimental effect of time delayson performance (e.g., Rahmandad et al., 2009),industry-wide time delays enhance firm and indus-try performance in simulations with two interdepen-dent rivals. Figure 4 illustrates these findings for



both conditions by showing a focal firm’s averagelevel of competitive activity and profits calculatedover 5,000 end-of-simulation values (t= 1,000).

The shape of the profit curve for interdepen-dent rivalry is markedly different from the curveobtained by assuming an independent firm becausetime delays cause market opportunities to becomeless contested (cf., Lenox et al., 2010: 123).Although firms explore fewer opportunities, as inthe independent firm scenario, firms also engageless in head-to-head competition, which increasesthe profits that accrue in the industry. There-fore, as long as time delays are industry-wide,both competitors benefit. As delays grow verylong, however, the positive performance effecteventually diminishes because customers have anone-of-the-above alternative (i.e., x𝜙 > 0) that theycan choose if industry players invest minimally incompetitive actions.

An examination of firm-specific time delays fur-ther clarifies our main results for time delaysand performance. Figure 5 depicts the impact ofthe focal firm’s execution delays on the firm’sperformance relative to its interdependent rival(Figure 5(a)) and the dyad’s aggregated profits(Figure 5(b)). Each line represents different levelsof the rival firm’s execution delay, ranging from 0(light grey) to 10 (black) in increments of two. Con-sistent with prior literature, the results demonstratethat slow competitors perform worse than theirfast counterparts (Chen et al., 2010a). However,a previously unrecognized nuance also emerges.Firm-specific time delays have a positive effect onindustry-level profits, up to a point, in line with theresults concerning industry-wide time delays (seeFigure 4). The turning point in profits is reachedsooner if the rival firm’s execution delay or theattractiveness of the none-of-the-above alternative,x𝜙, is large.

Note that execution delays are harmful—notbecause they decelerate cash flows or increasethe hazard of preemption (Hawk et al., 2013;Pacheco-de-Almeida, Hawk, and Yeung, 2015)but through the mechanism of reinforcementlearning—as these delays lead to suboptimal levelsof investment in competitive actions. Time delaysin decision-making feedback can complicate learn-ing in ways that make it more difficult for decisionmakers to ascertain how choices are related tooutcomes (Moxnes, 1998; Sterman, 1989, 2001). Inparticular, because firms must make up-front com-mitments to competitive actions, execution delays

cause managers to underestimate systematicallythe potential gains from investing in competitiveactions, thereby decreasing managers’ motivationto invest. As a result, a firm underutilizes avail-able market opportunities. However, the samemechanism causes industry profits to increase.As a focal firm invests less in competitive actionsthat attract the rival’s customers, competitiondeintensifies. The resulting increase in the rivalfirm’s performance tends to be larger than the focalfirm’s performance loss, until the delays grow verylarge.

Time delays and performance heterogeneity

Because performance feedback does not unambigu-ously direct firms toward a decision, time delaysin performance feedback may foster heterogene-ity in managers’ choices and in firms’ perfor-mance (e.g., King, 2007; Rahmandad, 2008; Rah-mandad et al., 2009). We contextualize this gen-eral assertion in interfirm rivalry by specifyingthe mechanism through which rivalry-related timedelays create heterogeneity (Harris, Johnson, andSouder, 2013).

Figure 6 illustrates how industry-wide changesin time delays (market reaction lag) affect the het-erogeneity produced by the process of competitiveinteraction. For each value of the time delay param-eter, we calculate the average difference betweenrivals in terms of competitive activity and profitsover 5,000 end-of-simulation values. For bench-marking purposes, we also calculate these indicesby randomly pairing 10,000 independent firm sim-ulations (i.e., we analyze 5,000 pairs of independentfirms). The results show that time delays contributeto heterogeneity in competitive activity, both acrosspaired independent firms and between direct rivals.The effect of time delays on performance hetero-geneity is even larger because time delays deinten-sify competition, which amplifies the performancedifference produced by a given disparity betweentwo firms’ levels of activity.

Although the pattern of results is similarregardless of our assumption about competi-tive interdependence vs. independence, there isan important difference in the mechanism thatproduces the results. For independent firms, het-erogeneity emerges simply because time delaysincrease firm-level oscillation in competitive activ-ity and profits. A large firm-level oscillation impliesthat, at any given point in time, two randomly paired


518 J. Luoma et al.

Competitive activity

0

5

10

15

20

25

30C

ompe

titiv

e ac

tivity

(xi

)

Market reaction lag (�m)

Dyadic rivalry Independent firm

0 1 2 3 4 5 6 7 8 9 10

(a)Profit

Market reaction lag (�m)

0

5

10

15

20

25

Profi

t (πi

)

Dyadic rivalry Independent firm

0 1 2 3 4 5 6 7 8 9 10

(b)

Figure 4. The effect of industry-wide time delays (market reaction lag) on the focal firm’s (a) competitive activity and(b) profit

(a) (b)Relative profit Dyad profit6

–30

–20

–10

0

10

20

30

40

0 1 2 3 4 5 6 7 8 9 10

Rel

ativ

e pr

ofit (

π i–π j)

Focal firm execution delay (τe)30

32

34

36

38

40

42D

yad

profi

t (π i+

π j)

Focal firm execution delay (τe)

Rival firm execution delay = 0




0 1 2 3 4 5 6 7 8 9 10

Figure 5. The effect of firm-specific time delays (execution delay) on (a) the focal firm’s profit relative to its rival and(b) the dyad’s total profita

a Note that the x-axis crosses the y-axis at 30.

firms are unlikely to behave and perform simi-larly. This implication is consistent with howthe existing literature has linked time delays toheterogeneity (e.g., Rahmandad, 2008; Rahmandadet al., 2009). In contrast, the primary mechanismthrough which time delays foster heterogeneityin a rivalrous setting is that industry-wide timedelays amplify small differences in rivals’ decisionmaking, causing interdependent rivals to followdivergent behavioral trajectories and consequentlyexperience differential levels of performance.

This pattern is illustrated in Figure 7, whichshows the correlation between (direct) rivals’levels of competitive activity and profits over 5,000end-of-simulation values. The results show thatincreases in market reaction lags are associatedwith increasingly negative correlations between

competitors’ levels of competitive activity andprofits. In other words, time delays cause rivals’behavioral and performance trajectories to diverge.

The following stylized scenario illustrates thelogic underlying this pattern. Consider a decisionpoint (A) for two equally active rivals. The focalfirm increases competitive activity, and its rivaldecides to decrease its level of competitive activ-ity. As a result, the focal firm will generally experi-ence a performance increase, and the rival’s perfor-mance will probably decline. This feedback encour-ages both firms to increase competitive activity inthe subsequent decision-making round (B). How-ever, time delays cause the performance effects ofthe firms’ decisions to carry over to subsequentrounds (i.e., B, C). The focal firm’s increased invest-ment in round A will continue to increase that firm’s



Market reaction lag (τm) Market reaction lag (τm)

(a)Competitive activity

(b)Profit

0

1

2

3

4

5

6

7

8

9|x

i - x

j|

Dyadic rivalry Independent firms (paired)

02468

101214161820

0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10

|πi-π

j|

Dyadic rivalry Independent firms (paired)

Figure 6. The contribution of industry-wide time delays (market reaction lag) to heterogeneity in competitive behaviorand performance. (a) Competitive activity and (b) profit

–1–0.9–0.8–0.7–0.6–0.5–0.4–0.3–0.2–0.1

0

0 1 2 3 4 5 6 7 8 9 10

Cor

rela

tion

Market reaction lag (τm)

Competitive activity Profit

Figure 7. Effect of industry-wide time delays (marketreaction lag) on the divergence of rivals’ behavior and

performance

performance, and the rival’s performance will con-tinue to decline. Similarly, the rival’s decision toreduce costs in round A will continue to affect thatfirm’s sales negatively while increasing the focalfirm’s performance. Therefore, while both firmsincrease their competitive activity in round B, theyare likely to experience qualitatively different per-formance effects. The focal firm, encouraged bypositive performance feedback, will be more likelyto increase competitive activity in round C, whereasthe rival firm, discouraged by negative performancefeedback, will be more likely to reduce costs again.In this way, the behavioral trajectories of the rivalstend to diverge, resulting in performance hetero-geneity between rivals.

The emerging differences in the firms’ levels ofcompetitive activity do not relate to ex ante differ-ences in known, competitor-specific drivers of com-petitive aggressiveness (e.g., Chen et al., 2010a;Ferrier, 2001). Heterogeneity emerges because the

time delays amplify small differences in the earlierdecisions of competitors. Essentially, heterogeneityemerges endogenously from the “history” of com-petition, as time delays amplify the amount of het-erogeneity that history produces.

Robustness analysis

We performed extensive robustness tests to gaugethe boundary conditions of our results, running themodel 500 times per time delay parameter value(and more, if needed) for every alternative modelspecification (see Table 2). First, holding otherparameters at their default values, we identified atheoretically meaningful lower and upper bound-ary for parameters affecting the costs and rewardsof competitive actions, that is, S, x𝜙, Ci and 𝛼

(parameter x0 was set at 10). For example, a nat-ural lower boundary for the cost of competitiveactions, Ci is zero. The upper boundary of Ci issuch that the xi

NE is zero actions per firm per timeperiod. A similar procedure yielded the lower andupper bounds on 𝛼 and x𝜙. We reran the analyses atseveral points between the extreme values. For themarket-size parameter S, we simply reran the anal-yses with a market size much larger than the defaultvalue.

We found the positive performance impact ofindustry-wide time delays to diminish with largevalues of Ci and x𝜙 and with small values of 𝛼 andS. The logic here is that if circumstances encouragelow levels of competitive activity relative to theattractiveness of the none-of-the-above alternative,then time delays cause customers—through firms’reduced competitive activity—to be attracted tothat option, causing the decline in competing firms’


520 J. Luoma et al.

Table 2. Summary of robustness analyses

Parameter orvariable Boundaries Tested values Robust rangea Observations

Rewards and costs of competitive actionsb

x𝜙 Min= 0Max= 100c

{0.001, 0.01, 1, 12, 24,50, 90, 100}

{0.001, 0.01, 1, 12, 24,50}

All results are robust forsufficiently low values.However, combinations ofa large x𝜙 (e.g., 50) and along rival firm executiondelay (e.g., 𝜏e = 5) causesincreases in the focalfirm’s execution delays toaffect industry-levelprofits negatively

𝛼 Min= 0 {0.1, 0.3, 0.5, 0.7, 1, 1.5,1.7, 1.9}

{0.7, 1, 1.5, 1.7, 1.9} All results are robust forsufficiently high valuesMax= 1.937d

Ci Min= 0 {0.05, 0.1, 0.25, 0.5, 1,1.5, 2, 3, 4, 5, 6, 7, 8}

{0.05, 0.1, 0.25, 0.5, 1,1.5}

All results are robust forsufficiently low valuesMax= 8.33c

S Min= 12c {12.5, 13, 15, 20, 30, 40,50, 60, 70, 80, 90, 100,1,000}

{60, 70, 80, 90, 100,1,000}

All results are robust forsufficiently high valuesMax=N/A

Decision making about competitive actionsw Min= 0 {0.0, 0.1, … , 0.9} {0.0, 0.1, … , 0.9} All results are robust within

the tested range ofparameter values

Max= 1

L(t) — {Poisson(1), Poisson(3),Poisson(5),Beta(0.5,0.5),Beta(1,3), Beta(2,2),Beta(2,5), Beta(5,1),Unif(1), Unif(3)}

{Poisson(1),Beta(0.5,0.5),Beta(1,3), Beta(2,2),Beta(2,5), Unif(1)}

All results are robust withinthe tested range ofdistributions that areskewed to the left(including a uniformdistribution) if the averagestep size is sufficientlysmall

Simulation controlFinal time {1,000, 2,500, 5,000,

10,000}{1,000, 2,500, 5,000,

10,000}All results are robust within

the tested range ofparameter values

a Robust range refers to the set of model specifications where all of the paper’s results remain qualitatively unaltered unless specifiedotherwise in the table. Besides changing the model parameters, we also tested alternative operationalizations of industry-wide andfirm-specific time delays. Industry-wide time delays were modeled using market reaction lags as well as industry-wide symmetricaland asymmetrical execution delays. Firm-specific time delays were modeled using symmetrical and asymmetrical execution delays. Inthe robust range, all results hold using all tested operationalizations of the time delays. The robust range for individual results can bebroader.b The firms’ initial levels of competitive activity, x0, were set at the Nash equilibrium values in all robustness analyses.c Calculated analytically using Mathematica: finding parameter values for which Nash equilibrium is 0.d Calculated numerically. On each step, the firms take turns to maximize their profit with respect to the current position of the rival.Implemented using scipy/fminbound algorithm in Python. As 𝛼 increases, firms’ profits in the Nash equilibrium decrease, eventuallyreaching 0.

profits. Thus, our claims about the positive perfor-mance impact of industry-wide time delays are validas long as competition occurs among incumbentsrather than with entrants or substitutes (and as longas time delays are not extremely large relative to thedecision-making interval, T). These assumptionshold especially in mature, oligopolistic industries.

Similarly, the contribution of time delays tointerfirm (performance) heterogeneity diminishesif the model parameters encourage low levels ofcompetitive activity. Because zero actions per timeperiod represent a lower boundary for firms’ levelsof competitive activity, there is no “room” forthe firm-level variance of competitive activity to



increase. This limits the amount of inter-rival (per-formance) heterogeneity produced by time delays.

Second, we tested different assumptions aboutfirms’ decision-making behavior. In particular, wechanged the average magnitude of the changes infirms’ levels of competitive activity as well as theshape of the distribution of Li. The results were sim-ilar for all of the tested left-skewed distributions ofLi (i.e., when incremental changes in competitiveactivity are assumed to be more likely than dras-tic changes). More important, we found that theeffect of time delays on inter-rival (performance)heterogeneity is diminished by large average val-ues of Li. The logic behind this result is similar tothat identified in the context of parameters impact-ing the rewards and costs of competitive actions. Ifthe magnitude of changes in firms’ levels of com-petitive activity is large, there is little additionalroom for firm-level variance in competitive behav-ior and, consequently, little room for inter-rival (per-formance) heterogeneity to increase. Theoretically,the findings imply that the model predictions aremost relevant in contexts where firms take numer-ous competitive actions in each time period andadjust their competitive behaviors incrementally inview of their experience. We also varied the firms’reliance on historical vs. recent performance feed-back across the full theoretically meaningful range(i.e., 0≤w< 1). We found that smaller (larger) val-ues of w tend to increase (decrease) the amountof heterogeneity produced by time delays. Relianceon historical feedback reinforces path dependency,which increases the amount of heterogeneity cre-ated in the absence of time delays. This effect inturn reduces how much additional heterogeneity isproduced by time delays. Finally, we examined analternative, asymmetrical operationalization of theexecution delay (Equation 3) in which decreasesin competitive activity take effect instantaneously.Because asymmetrical execution delays cause onlyfeedback about increases in competitive activity tobe delayed, asymmetrical execution delays influ-ence the model dynamics less than execution delayswith a symmetrical form; however, the qualitativeconclusions remain the same.

DISCUSSION

Our study examines time delays as an antecedentof industry profitability and performance hetero-geneity among rivals. To this end, we develop a

computational model that is built around the notionthat rivalrous relationships evolve over time throughrepeated encounters (Kilduff et al., 2010). Our com-putational experiments examine how time delays—in rivals’ implementation of, or market responsesto, competitive actions—influence learning andthereby patterns of rivalry and firm performance.Scholars in the field of competitive dynamics andlearning usually assert that time delays diminish afirm’s performance and that competitive speed andprompt feedback are desirable. Although our modelis capable of reproducing these results, the modelalso highlights previously unrecognized implica-tions and boundary conditions on these claims.

First, our study finds that a firm that is slow toimplement competitive actions tends to be outper-formed by its faster rival. Although this result isunlikely to surprise the reader, the mechanism isless known in the competitive dynamics literature.Our results emphasize that certain benefits of speedmaterialize over time through learning, which isdistinct from more widely recognized benefits ofspeed such as minimizing the risk of preemption(Hawk et al., 2013) and reducing the competitor’sability to build protective fences against attacks(Boyd and Bresser, 2008). We find that a firm’squicker competitive maneuvering may enhance itsrelative performance by enabling superior learningfrom more accurate performance feedback, leadingto more optimal utilization of market opportunities.

Second, we consider the industry-wide perfor-mance implications of a focal firm’s executiondelays. According to our results, execution delaysthat are specific to one competitor can have a netpositive effect on industry profits. An alternativereading of this result is that a focal firm’s effortsto increase performance, through greater executionspeed, triggers an even greater downfall in its rivalfirm’s performance, resulting in a net negativeeffect on industry attractiveness. Prompt feedback,implied by brief execution delays, causes a firm toperceive greater profit in engaging in direct compe-tition with its rival. This implies that competitors’parallel investments in competitive maneuveringcapabilities are a negative-sum game, leadingto increasing levels of competitive intensity anddeclining profits.

More generally, industry-wide time delays (i.e.,time delays that concern all rivals) result in higherindustry profits. Industry-wide time delays affectlearning in ways that act as a barrier to escalatingprofit-destroying competition. This effect enhances


522 J. Luoma et al.

the performance of competing firms as rivals com-pete less aggressively for the same market oppor-tunities. This result stands in stark contrast to theexisting literature on time delays in feedback-basedlearning, where the opposite performance effect isusually predicted (Denrell et al., 2004). However,the results differ simply because previous studieshave not incorporated competitive interdependenceinto their analyses of the performance effect of timedelays. In essence, we conceptualize time delays asan impediment to direct competition among incum-bents that may increase their profits and the indus-try’s attractiveness.

Third, our paper offers insights into how timedelays foster performance heterogeneity in rivalryamong ex ante identical firms. Our findings areconsistent with the existing learning literatureindicating that time delays increase the difficultyof linking competitive actions to performanceoutcomes (Rahmandad, 2008). As a result, timedelays lead to oscillation in an independent firm’slevel of competitive activity. In the context of twointerdependent rivals, the mechanism producingheterogeneity is somewhat different. Time delaysfoster divergence in competitors’ behavioralpatterns. Time delays amplify small inter-rivaldifferences in decision making about competitiveactions, leading one firm to invest aggressively asits rival remains competitively inactive. Thus, wereveal previously unspecified complexity in howtime delays may be related to performance hetero-geneity. By doing so, we establish the constructmore firmly alongside other behavioral explana-tions of heterogeneity (e.g., Camerer and Lovallo,1999; Lenox et al., 2006). These results also con-tribute to the competitive dynamics literature, as wedemonstrate that the factors that influence competi-tors’ learning from previous encounters (e.g., timedelays) may shape the amount of interfirm hetero-geneity that competition endogenously produces.

Beyond exploring the role of time in rivalrythrough a learning perspective, our study addressesa more general research gap noted by Chen andMiller (2012), as we explore how the interde-pendence of competitors’ actions and responsesgives rise to subsequent competitive behaviors(ibid.: 173). Specifically, the study explains how“longer-term patterns of interfirm rivalry” (Marcelet al., 2010: 131) can emerge from inter-temporallyconnected decisions about competitive actions.This undertaking required an examination ofhow learning from previous decisions and their

outcomes affect subsequent decisions about com-petitive actions. The interlinking of decisionsover time renders competition a path-dependentphenomenon. The range of potential outcomes ofsuch a process is difficult to predict using intuitionand verbal reasoning (Sastry, 1997), which is whywe employed computational modeling. In doingso, we point to a new methodological direction forcompetitive dynamics research, which has reliedprimarily on verbal reasoning and large competitiveaction datasets to develop and test theories (Chenand Miller, 2012).

Directions for future research

Our findings lead to several new directions forstrategic management researchers to pursue. Onefruitful direction for future research is to examineempirically the relationship among industry-widetime delays, the intensity of rivalry, and incum-bents’ profits. An obvious hypothesis emergingfrom our findings is that time delays could predictprofitability differences across industries (cf.,Lenox et al., 2010). It should be noted, however,that competitive forces from outside the industryultimately limit the positive performance impactof time delays (as is true of any mechanism thatlimits competition among incumbents). Moreover,because time delays simultaneously increaseinter-rival performance differences, the profitincrease is unlikely to be equal across competitors.More aggressive competitors will probably garnergreater gains in profits if feedback is delayed.A multi-industry sample (King and Zeithaml,2001) that captures significant industry differencesin time delays (e.g., software firms or Internetpublishing vs. pharmaceutical or aerospace) wouldfacilitate deeper empirical insights into these kindsof performance implications of time delays.

Insights that time delays contribute to hetero-geneity in competitive behavior and performancealso merit further empirical research. Previouscompetitive dynamics research implies that dif-ferences between rivals’ levels of competitiveaggressiveness might emerge in a path-dependentmanner, as past performance affects the futurecompetitive aggressiveness of firms (e.g., Ferrier,2001). Our results complement these findings byspecifying the conditions that foster interfirm het-erogeneity in competitive behavior. For example,we should observe greater heterogeneity amongfirms in their use of major product introductions



but little variation between rivals in their use ofprice changes. Ultimately, industries with delayedfeedback on central competitive actions are likelyto display greater performance heterogeneity thanindustries with prompt feedback about the mostimportant competitive moves. Our robustnessanalyses indicate that these predictions are likelyto be most relevant in contexts where the volumeof actions taken by competitors is large relativeto the magnitude of changes in those volumes.Along with observational studies, experimentsand experiential simulations are valuable researchmethods for investigating the effects of time delaysand learning on rivals’ development of competitivestrategies (Chen, Lin, and Michel, 2010b).

In addition to testing empirical predictions thatemerge from our model, an equally important pathof inquiry involves expanding the use of compu-tational modeling to study rivalry. For example,although our model assumes that firms are primarilyconcerned with improving their own performance,computational experiments could be designed toexamine assumptions about retaliatory behavior(e.g., Chen, Su, and Tsai, 2007) and imitation (e.g.,Hsieh et al., 2014). Time delays are likely to influ-ence both the probability of retaliatory behavior andmanagers’ evaluations of the attractiveness of imi-tating a strategy employed by rivals. Researcherscould also build more complex models of com-petitors’ learning and reasoning about rival firms’competitive actions and specify the effect of timedelays under different assumptions about rationality(Rahmandad et al., 2009).

In addition to examining the effect of time delayson rivalry, computational modeling approachescould be used to address other questions of interestto competitive dynamics scholars. Possibilitiesinclude modeling the emergence and consequencesof competitive asymmetry (e.g., Chen, 1996) andmultimarket contact (e.g., Baum and Korn, 1999)as a result of firms’ evaluation of feedback fromrepeated interactions. Likewise, following Lenoxet al. (2006), interesting insights may emerge iffuture scholars introduce entry-exit dynamics andrivalry among incumbent firms in one model.

Other simulation techniques can be used toaddress important research questions pertain-ing to competitive dynamics. For example, systemdynamics models treat variables such as competitiveactivity as continuous. However, in some situations,changes in competitive activity may occur in dis-crete “lumps” that may subsequently preempt a rival

firm’s competitive actions (Bromiley, Papenhausen,and Borchert, 2002). Discrete-event simulationmodels might be used to capture such dynam-ics. Agent-based models, in turn, could capturedemand-side heterogeneity with greater granularity.For example, a history of competitive interactionmight cause a firm to attract price-sensitive cus-tomers as its rival attracts customers that are drawnto prestigious brands. Such differences could haveinteresting implications for subsequent competitivedecision making by generating heterogeneity inhow customers respond to competitors’ actions.

Finally, we hope this study motivates compu-tational strategy scholars, who frequently assumefirms to be unaffected by the actions of rivals, toexamine systematically interactions between com-petition and their focal construct. As this paperdemonstrates, the assumption of firm centricity canlead to very different conclusions about the behav-ioral dynamics of strategy compared with situationsthat involve two or more interdependent rivals.

ACKNOWLEDGEMENTS

The authors wish to acknowledge the editor, Profes-sor Richard Bettis, and two anonymous reviewersfor their excellent guidance throughout the reviewprocess. Thoughtful insights from Ming-Jer Chen,Tomi Laamanen, and Michael Lenox also helpedimprove the paper significantly. Jaakko Aspara,Tomas Falk, Johanna Frösén, Matti Jaakkola,Juha-Antti Lamberg, and Jeremy Marcel provideduseful comments in various stages of the project,as did the conference and seminar audiences atthe Academy of Management Annual Meeting,2012, European Academy of Management AnnualMeeting, 2014, and the University of Kentucky,2015. Jukka Luoma wishes to acknowledge thefinancial support of the Finnish Funding Agencyfor Innovation, Finnish Foundation for EconomicEducation, Fulbright Center Finland, HSE Foun-dation, Jenny and Antti Wihuri Foundation andKAUTE Foundation. Professor King’s researchwas funded by the Ray Hunt Faculty FellowshipFund at the McIntire School of Commerce.

REFERENCES

Ancona DG, Goodman PS, Lawrence BS, Tushman ML.2001. Time: a new research lens. Academy of Manage-ment Review 26(4): 645–663.


524 J. Luoma et al.

Aspara J, Pajunen K, Tikkanen H, Tainio R. 2014. Explain-ing corporate short-termism: self-reinforcing processesand biases among investors, the media and corporatemanagers. Socio-Economic Review 12(4): 667–693.

Atkins PWB, Wood RE, Rutgers PJ. 2002. The effects offeedback format on dynamic decision making. Orga-nizational Behavior and Human Decision Processes88(2): 587–604.

Baum J, Korn H. 1999. Dynamics of dyadic competi-tive interaction. Strategic Management Journal 20(3):251–278.

Becker MC. 2004. Organizational routines: a review ofthe literature. Industrial and Corporate Change 13(4):643–678.

Boyd JL, Bresser RKF. 2008. Performance implications ofdelayed competitive responses: evidence from the U.S.retail industry. Strategic Management Journal 29(10):1077–1096.

Bridoux F, Smith KG, Grimm CM. 2013. The managementof resources: temporal effects of different types ofactions on performance. Journal of Management 39(4):928–957.

Bromiley P, Papenhausen C, Borchert P. 2002. Whydo gas prices vary, or towards understanding themicro-structure of competition. Managerial and Deci-sion Economics 23(4–5): 171–186.

Camerer C, Lovallo D. 1999. Overconfidence and excessentry: an experimental approach. American EconomicReview 89(1): 306–318.

Chatain O, Zemsky P. 2011. Value creation and valuecapture with frictions. Strategic Management Journal32(11): 1206–1231.

Chen EL, Katila R, McDonald R, Eisenhardt KM. 2010a.Life in the fast lane: origins of competitive interactionin new vs. established markets. Strategic ManagementJournal 31(13): 1527–1547.

Chen M-J. 1996. Competitor analysis and interfirm rivalry:toward a theoretical integration. Academy of Manage-ment Review 21(1): 100–134.

Chen M-J, Hambrick DC. 1995. Speed, stealth, and selec-tive attack: how small firms differ from large firms incompetitive behavior. Academy of Management Jour-nal 38(2): 453–482.

Chen M-J, Lin H-C, Michel JG. 2010b. Navigating ina hypercompetitive environment: the roles of actionaggressiveness and TMT integration. Strategic Man-agement Journal 31(13): 1410–1430.

Chen M-J, Miller D. 2012. Competitive dynamics: themes,trends, and a prospective research platform. Academy ofManagement Annals 6(1): 135–210.

Chen M-J, Su K, Tsai W. 2007. Competitive tension: theawareness-motivation-capability perspective. Academyof Management Journal 50(1): 101–118.

Cyert RM, March JG. 1992. A Behavioral Theory of theFirm (2nd edn). Prentice-Hall: Englewood Cliffs, NJ.

Denrell J, Fang C, Levinthal DA. 2004. From T-mazesto labyrinths: learning from model-based feedback.Management Science 50(10): 1366–1378.

Derfus PJ, Maggitti PG, Grimm CM, Smith KG. 2008.The Red Queen effect: competitive actions and firmperformance. Academy of Management Journal 51(1):61–80.

Diehl E, Sterman JD. 1995. Effects of feedback complexityon dynamic decision making. Organizational Behaviorand Human Decision Processes 62(2): 198–215.

Eisenhardt KM. 1989. Making fast strategic decisions inhigh-velocity environments. Academy of ManagementJournal 32(3): 543–576.

Farris P, Verbeke W, Mdickson P, Van Nierop E. 1998.Path dependencies and the long-term effects of rou-tinized marketing decisions. Marketing Letters 9(3):247–268.

Ferrier WJ. 2001. Navigating the competitive landscape:the drivers and consequences of competitive aggres-siveness. Academy of Management Journal 44(4):858–877.

Ferrier WJ, Smith KG, Grimm CM. 1999. The role ofcompetitive action in market share erosion and indus-try dethronement: a study of industry leaders andchallengers. Academy of Management Journal 42(4):372–388.

Gersick CJG. 1994. Pacing strategic change: the case of anew venture. Academy of Management Journal 37(1):9–45.

Greve HR. 2002. Sticky aspirations: organizational timeperspective and competitiveness. Organization Science13(1): 1–17.

Hambrick DC, Cho TS, Chen M-J. 1996. The influence oftop management team heterogeneity on firms’ compet-itive moves. Administrative Science Quarterly 41(4):659–684.

Harris JD, Johnson SG, Souder D. 2013. Model-theoreticknowledge accumulation: the case of agency theory andincentive alignment. Academy of Management Review38(3): 442–454.

Hawk A, Pacheco-De-Almeida G, Yeung B. 2013.Fast-mover advantages: speed capabilities and entryinto the emerging submarket of Atlantic Basin LNG.Strategic Management Journal 34(13): 1531–1550.

Horsky D. 1990. A diffusion model incorporating productbenefits, price, income and information. MarketingScience 9(4): 342–365.

Hotelling H. 1929. Stability in competition. EconomicJournal 39(153): 41–57.

Hsieh K-Y, Tsai W, Chen M-J. 2014. If they can do it, whynot us? Competitors as reference points for justifyingescalation of commitment. Academy of ManagementJournal 58(1): 38–58.

Katila R, Chen EL. 2008. Effects of search timing oninnovation: the value of not being in sync with rivals.Administrative Science Quarterly 53(4): 593–625.

Kilduff GJ, Elfenbein HA, Staw BM. 2010. The psychol-ogy of rivalry: a relationally dependent analysis ofcompetition. Academy of Management Journal 53(5):943–969.

King AW. 2007. Disentangling interfirm and intrafirmcausal ambiguity: a conceptual model of causal ambi-guity and sustainable competitive advantage. Academyof Management Review 32(1): 156–178.

King AW, Zeithaml CP. 2001. Competencies and firmperformance: examining the causal ambiguity paradox.Strategic Management Journal 22(1): 75–99.



Laamanen T, Keil T. 2008. Performance of serial acquir-ers: toward an acquisition program perspective. Strate-gic Management Journal 29(6): 663–672.

Lam CF, DeRue DS, Karam EP, Hollenbeck JR. 2011. Theimpact of feedback frequency on learning and task per-formance: challenging the ‘more is better’ assumption.Organizational Behavior and Human Decision Pro-cesses 116(2): 217–228.

Lamberg JA, Tikkanen H, Nokelainen T, Suur-InkeroinenH. 2009. Competitive dynamics, strategic consistencyand organizational survival. Strategic ManagementJournal 30(1): 45–60.

Lenox MJ, Rockart SF, Lewin AY. 2006. Interdependency,competition, and the distribution of firm and industryprofits. Management Science 52(5): 757–772.

Lenox MJ, Rockart SF, Lewin AY. 2010. Does inter-dependency affect firm and industry profitability? Anempirical test. Strategic Management Journal 31(2):121–139.

Levinthal DA, March JG. 1993. The myopia of learning.Strategic Management Journal 14(S2): 95–112.

Lippman SA, Rumelt RP. 1982. Uncertain imitability: ananalysis of interfirm differences in efficiency undercompetition. Bell Journal of Economics 13(2):418–438.

Lurie NH, Swaminathan JM. 2009. Is timely informationalways better? The effect of feedback frequency ondecision making. Organizational Behavior and HumanDecision Processes 108(2): 315–329.

MacMillan I, McCaffery ML, Van Wijk G. 1985. Com-petitors’ responses to easily imitated new products—exploring commercial banking product introductions.Strategic Management Journal 6(1): 75–86.

Marcel JJ, Barr PS, Duhaime IM. 2010. The influence ofexecutive cognition on competitive dynamics. StrategicManagement Journal 32(2): 115–138.

Miller D, Chen M-J. 1994. Sources and consequencesof competitive inertia: a study of the US airlineindustry. Administrative Science Quarterly 39(1):1–23.

Mosakowski E. 1997. Strategy making under causal ambi-guity: conceptual issues and empirical evidence. Orga-nization Science 8(4): 414–442.

Moxnes E. 1998. Not only the tragedy of the commons:misperceptions of bioeconomics. Management Science44(9): 1234–1248.

Nakanishi M, Cooper LG. 1974. Parameter estimation fora multiplicative competitive interaction model–least

squares approach. Journal of Marketing Research11(3): 303–311.

Pacheco-de-Almeida G, Hawk A, Yeung B. 2015. The rightspeed and its value. Strategic Management Journal36(2): 159–176.

Paich M, Sterman JD. 1993. Boom, bust, and failures tolearn in experimental markets. Management Science39(12): 1439–1458.

Rahmandad H. 2008. Effect of delays on complexity oforganizational learning. Management Science 54(7):1297–1312.

Rahmandad H, Repenning N, Sterman J. 2009. Effects offeedback delay on learning. System Dynamics Review25(4): 309–338.

Sastry MA. 1997. Problems and paradoxes in a modelof punctuated organizational change. AdministrativeScience Quarterly 42(2): 237–275.

Shanks DR, Pearson SM, Dickinson A. 1989. Temporalcontiguity and the judgement of causality by humansubjects. Quarterly Journal of Experimental Psychol-ogy Section B 41(2): 139–159.

Sloman SA. 1996. The empirical case for two systems ofreasoning. Psychological Bulletin 119(1): 3–22.

Smith KG, Grimm CM, Gannon MJ, Chen M-J. 1991.Organizational information processing, competitiveresponses, and performance in the U.S. domesticairline industry. Academy of Management Journal34(1): 60–85.

Srinivasan S, Vanhuele M, Pauwels K. 2010. Mind-set met-rics in market response models: an integrative approach.Journal of Marketing Research 47(4): 672–684.

Sterman JD. 1989. Misperceptions of feedback in dynamicdecision making. Organizational Behavior and HumanDecision Processes 43(3): 301–335.

Sterman JD. 2001. System dynamics modeling—tools forlearning in a complex world. California ManagementReview 43(4): 8–25.

Sterman JD, Henderson R, Beinhocker E, Newman L.2007. Getting big too fast: strategic dynamics withincreasing returns and bounded rationality. Manage-ment Science 53(4): 683–696.

Wasserman EA, Miller RR. 1997. What’s elementary aboutassociative learning? Annual Review of Psychology 48:573–607.

Young G, Smith KG, Grimm CM. 1996. ‘Austrian’ andindustrial organization perspectives on firm-level com-petitive activity and performance. Organization Science7(3): 243–254.


time delays, competitive interdependence and firm performance · 3 mcintire school of commerce,...

Documents