ntRelative Performance Evaluation and Target Setting

MARTIN HOLZHACKER
Erasmus University Rotterdam

MATTHIAS D. MAHLENDORF
WHU – Otto Beisheim School of Management

MICHAL MATĚJKA
W.P. Carey School of Business, Arizona State University

Draft – Please do not cite or circulate without permission of the authors

March 2013

2

Relative Performance Evaluation and Target Setting
ABSTRACT
It is well known that compensation contracts can improve risk sharing by incorporating information about peer performance ex post, i.e., once the performance of a unit as well as its peers is observed. In this study, we examine whether peer performance is also used ex ante,
i.e., when setting performance targets at the beginning of a period. We analyze data on 2008–
2010 performance targets from 354 units of a national agency responsible for reintegration of the long-term unemployed into the labor market. We find evidence that a target for a performance measure is strongly positively associated with past peer performance on the same measure as well as past peer performance on a different measure. Thus, performance targets are lower whenever past peer performance is weak on at least one of its dimensions.
We also find that the relative weights on different measures of peer performance are proportional to their capacity to filter out common risks. We conclude that using information about past peer performance improves risk sharing and alleviates adverse incentive effects associated with setting targets based on a unit’s own past performance.

Keywords: Relative performance evaluation; target setting; target ratcheting; multi-tasking.

Data availability: Data used in this study is not publicly available due to confidentiality agreements with participants of this study.

3

INTRODUCTION
Performance targets are an important component of management control systems by serving as a standard against which managerial performance is often evaluated at the end of the period. Thereby, targets largely determine bonuses, career prospects and promotion decisions to motivate managerial effort and retain employees. Given this prominent role of targets in business practice, it is important to understand how supervisors set performance targets to contract with managers.
Prior literature has distinguished between two major sources of information on which targets can be based: past performance and relative peer performance (Murphy 2001;
Milgrom and Roberts 1992, 233).1 As a result, two largely unrelated streams of literature have emerged. First, prior research on how to set budget targets ex ante (i.e., in the beginning of the period) has been concerned with how much weight firms place on past performance when setting targets. Single firm field evidence tends to suggest ratcheting2 targets which is equivalent to supervisors putting a heavy emphasis on past performance when setting bonus targets3 (Leone and Rock 2002; Bouwens and Kroos 2011). This, in turn, is costly as it will lead managers to attempt to influence their targets by engaging in potentially detrimental gaming behavior (e.g., effort reduction, accrual management and so forth). Contract theory suggests that it might not necessarily be optimal to fully exploit the information content of past performance in the first place. Instead, ex ante incentives can be strengthened by allowing managers to earn rents in future periods as well by not fully adjusting targets to past performance. Thus, such or other forms of long term commitment to preserve existing contractual arrangements might be a superior strategy (Laffont and Tirole 1993, 376). Studies using survey evidence from diverse samples of firms find support for the view that managers who achieve targets are likely to do so persistently over time (Indjejikian and Nanda 2002;
Indjejikian and Matĕjka 2006; Indjejikian et al. 2011). These findings suggest a more tempered use of past performance as target ratcheting appears to be limited to more moderate levels in their sample of firms. Second, contract theory predicts that managerial performance will be evaluated against peer performance to filter out common shocks which allows for
1

Milgrom and Roberts (1992) mention a third source which is time and motion studies. However, we do not consider it here because it is less appropriate to complex, non-repetitive tasks at the management level.
2
Ratcheting refers to the tendency to use past performance as a criterion in determining targets (Weitzman
1980).
3
Hereafter, we will refer to budget based targets set at the beginning of the period simply as “targets”.
Nevertheless, we acknowledge that targets can be set ex post as in the case of RPE.

4

higher precision at assessing managerial performance and more efficient contracting
(Holmstrom 1982; Matsumura and Shin 2006). Given the conjecture that any performance signal which is conditionally informative about managerial effort should be used in contracts
(Holmstrom 1979; Lambert 2001) supervisors are likely to exploit not only past peer performance, but also the information content of peer performance to set targets. To our knowledge, no prior study has investigated how and whether past peer performance is used to set performance targets. Given mixed evidence on relative performance evaluation (RPE)4 and that the majority of firms relies on budget targets based on prior performance in their annual incentive plans (Murphy 2001), it appears worthwhile to investigate whether relative performance evaluation can be implemented through targets. Thus, we fill this gap by providing field study evidence of how supervisors jointly incorporate past individual and peer performance signals into targets.
Drawing on the informativeness principle we predict the following. First, we hypothesize that past peer performance on a given performance measure will be incorporated into targets. Performance relative to target is a noisy signal of managerial effort period. The use of peer performance in target setting allows to filter out common exogenous shocks from performance signals. Placing positive weight on peer performance in target setting will allow to reward managerial performance more reliably with lower future targets. As favorable relative performers receive lower targets which leads them to earn higher future expected rents, ex ante incentives for current period managerial performance will be strengthened.
More specifically, incorporating peer performance into targets will decrease the cost of current period effort by reducing future target mark-ups resulting from excess performance relative to target and place an additional future cost on current period shirking.
Second, the informativeness principle implies that any performance signal that is conditionally informative about the manager’s effort beyond the information content of other measures should be weighted in performance evaluation according to its informational value
(Banker and Datar 1989). Consistent with this view, we argue that using multiple instead of a single measure of how managers performed relative to their peers allows to more reliably assess and reward managerial performance. To test this conjecture, we analyze whether peer performance on another performance measure will be incorporated into performance targets.
4

Prior research tests this conjecture using large scale executive compensation data, but finds mixed evidence for the use of relative performance evaluation (RPE) (e.g., Antle and Smith 1986, Gibbons and Murphy 1990,
Janakiraman et al. 1992). Especially the adequacy of empirical tests to detect RPE (Albuquerque 2009; Dikolli et al. 2012) and cross-sectional differences in RPE use (Aggarwal and Samwick 1999; Gong et al. 2011) have been subject to current debate.

5

Finally, the benefits of relative performance evaluation (RPE) largely depend on the capacity of the peer group to absorb common shocks (Albuquerque 2009; Gong et al. 2011; Dikolli et al. 2012). We predict cross-sectional differences in the use of the two performance signals varying based on their respective capacity to absorb common shocks. Thus, we analyze whether the ratio of weights assigned to the two measures of relative performance in target setting varies according to the relative information content of one relative performance measure over the other.
To analyze our hypotheses, we collect quantitative field data from 354 local branches, hereafter referred to as “employment centers”, of a national employment agency.5

6

Employment center managers are evaluated and compensated based on an aggregate of weighted performance measures against their respective targets which has been set at the beginning of the year. Our setting is particularly useful to study the use of RPE in target setting for three reasons. First, performance of employment center managers is largely driven by exogenous shocks (e.g., events in the global and local economy). Therefore, the benefits of using RPE in terms of filtering shocks from employment center performance can be considerable. Second, information on peer performance is abundantly available as
Employment Agency headquarters has composed internal peer groups and provides unrestricted access to performance reports and rankings through its accounting information system. Even though headquarters provides initial target proposals in a two-stage target setting process, supervisors and employment center managers have considerable discretion to use this information to alter target proposals. Access to data on both initial and final targets enables us to disentangle a system effect and the incremental effect of decentralized negotiations on targets. Third, access to data on two performance measures (“labor market integrations per customer” and “welfare payment savings”) allows us to study how multiple signals of peer performance are incorporated into targets.
Analysis of 974 employment center year observations during the years 2008 to 2010 indicates the following. Future targets regarding integrations per customer are positively related to peer performance on that same measure. Employment managers who outperform their peers receive relatively lower targets while employment center managers who underperform their peers receive a target mark-up. This suggests that incorporating peer
5

Due to a confidentiality agreement, the organization cannot be disclosed. We will hereafter refer to the organization as Employment Agency.
6
The employment centers are mainly responsible for training and reintegration of the long-term unemployed into the labor market.

6

performance into targets which are being used to compensate managers ensures that wellperforming managers will face lower targets in future periods, and therefore will earn future economic rents. Our finding is consistent across both measures of peer performance, integrations per customer and savings in welfare payments. This finding suggests that supervisors do not only use peer performance on a given measure, but may incorporate other available signals of performance which allows them to more accurately filter common noise from performance evaluations. Analysis allowing target response to peer performance signal to vary by high and low capacity7 of peer groups to filter common risk reveals the following.
For increases in peer group capacity to filter common risk from a certain performance measure, that measure of peer performance will receive a larger weight relative to other measures of peer performance. More specifically, target integrations per customer are more responsive to peer performance on that same measure relative to peer performance on the second measure savings in welfare payments if the extent of common risk on the performance measure integrations per customer (savings on welfare payments) borne by the peer group shifts from low (high) to high (low). Overall, our findings support the view that past peer performance is used as a signal to set future targets to the extent that it is conditionally informative about managerial performance and allows to reward managers more reliably for current effort by lowering their future targets.
Our study contributes to the literature in the following ways. First, we extend prior studies which suggest a more tempered use of past performance information for target setting by providing evidence that supervisors will not solely rely on past performance as performance signals to set budget targets. Instead, our findings suggest that supervisors exploit available information with respect to peer performance. If targets are based not only on past performance relative to target, supervisors decide to not fully exploit information associated with past performance. Instead, by relying on relative performance for target setting, supervisors avoid punishing high levels of current managerial performance through target ratcheting and thereby commit to existing contractual arrangements, that is well performing managers get to keep their rents persistently over time.

7

According to Holmstrom and Milgrom (1987), peer performance measures with high correlations to own performance have a higher ability to filter out common shocks. To capture the ability of a peer group to insulate a certain employment center from common uncertainty, we construct an optimal peer group for the given employment center based on the employment centers with the highest time series correlation on a given performance measure. We assume the extent of overlap between actual and optimal peer group composition to proxy for the extent of common risk that the employment center shares with its actual peer group.

7

Second, our study contributes to furthering the understanding of how RPE is implemented in a multi-period setting. Given a common field setting where performance evaluation is limited to evaluation of actual performance against a budgeted target, using current period peer performance to compensate managers in the same period might not be feasible. Instead, incorporating peer performance into future targets might effectively serve as an insurance against common shocks that allows to more reliably reward well performing managers with higher rents through lower targets in future periods.
Third, our study is the first to show how multiple signals of performance are weighted to implement RPE, which is consistent with the widely-held view that any signal of informational value should be weighted accordingly in performance evaluation. This is particularly important given the fact the most managers are evaluated based on a weighted aggregate of performance measures which are all subject to various common and idiosyncratic risks.
The remainder of the paper is structured as follows. In section II, we revisit the theory and derive the hypotheses for our study. Section III explains the research setting chosen for our study, while section IV provides more information with respect to the research design of our study. Section V presents the empirical models and empirical results. Finally, section V concludes our study.

THEORY AND HYPOTHESES
Targets play an important role in performance evaluation of mangers. Performance against target is widely used as a performance signal which is seen to be informative about managerial performance and compensated accordingly to elicit high levels thereof.
Specifically, performance falling short of target is typically associated with a smaller or no bonus at all, bleak promotions prospects and loss of reputation. In turn, performance exceeding target may yield substantial bonus payments and facilitate career advancement. As a result, targets levels determine manager’s marginal returns to effort as well as the compensation risk borne by the managers, which are both important parameters of managerial incentives. Jointly with choice of performance measures and pay-performance relationship, target levels calibrate the manager’s compensation function.

8

In spite of targets’ relevance to contract design, contract theory is mostly concerned with how to structure the contract of a manager to elicit high performance, providing little explicit advice on how targets should be set. Milgrom and Roberts (1992) distinguish between three sources of information used for target setting. First, targets may be set based on time and motion studies which are rather applicable to highly repetitive tasks. Second, targets may be set ex post (i.e., at the end of the period) based on the realized performance of a group of peers. Third, targets may be set ex ante (i.e., in the beginning of a period) based on past performance of an individual manager. Accounting studies have mostly been concerned with either one of the latter two. Murphy’s (2001) seminal paper is among the few which discusses the cost and benefits of using past performance vis-à-vis peer performance to set targets for performance evaluation purposes. He argues that the choice of a target setting method depends on how costly the respective method of estimating the target is (e.g., in terms of data collection effort and timeliness of information), how accurately noise (i.e., exogenous shocks affecting performance) can be estimated (i.e., to be incorporated into the standard), and the extent to which managers can have discretion to influence their future target (e.g., by withholding effort).
Using past performance for target setting is less costly in terms of data collection effort because data on an individual manager’s performance is usually readily available from a firm’s accounting information system in the beginning of the period when targets are set. Yet, setting targets based on past performance deteriorates manager’s ex ante incentives to exert effort, a dynamic incentive problem which is known as the “ratchet effect” (Weitzman 1980).
That is, if a manager anticipates that sending a signal of high effort by outperforming her target will prove costly as her target will be ratcheted up, she will withhold effort in order to meet but not beat the target and maintain existing target levels. Single firm field studies find supporting evidence for this behavior. Leone and Rock (2002) document income-decreasing accrual management in response to ratcheting budgets at a business unit of a large North
American corporation. Bouwens and Kroos (2011) study the sales performance of electronics store managers and find that managers with favorable year-to-date budget performance withhold sales efforts.
Another significant drawback in using past performance to set targets is that it is a noisy signal of managerial effort. Avoiding risk exposure of managerial performance evaluation has been shown to an important concern in target setting (Indjejikian and Nanda 2002; Bol et al.
2010). Thus, firms operating in environment with high volatility in earnings tend to place less

9

weight on past performance (or negotiations for that matter8) to set targets for evaluation purposes because it is more costly to contract on a noisy measure (Murphy 2001). Contract theory suggests that the benefits from using peer performance increase with the extent of common risk in the environment as relative performance evaluation (RPE) allows to filter out the shocks that are common to peer group members from the manager’s performance measure
(Holmstrom 1982). The informativeness principle established by Holmstrom (1979) proposes that any performance measure that is informative about a manager’s actions should be included in performance evaluation. This implies that even though peer performance is not sensitive to a manager’s effort, it may still be a valuable performance measure in contracting.
In an environment characterized by high degrees of common risk, peer performance will be highly correlated to individual performance due to a common noise term. By assigning negative weight to peer performance in a manager’s performance contract, common shocks may by filtered out from a manager’s individual performance measure, which allows to more accurately assess and compensate managerial effort (Lambert 2001; Gong et al. 2011).
In spite of the prediction of relative performance evaluation hypothesis that firms or supervisors find it beneficial to evaluate managerial performance relative to their peers’, empirical evidence regarding whether relative performance is incorporated into managerial contracts has produced mixed findings (e.g., see Albuquerque 2009 for a review). A number of prior studies tend to suggest cross-sectional differences regarding the costs and benefits associated with the use of RPE (Aggarwal and Samwick 1999; Matsumura and Shin 2006;
Gong et al. 2011). In fact, when using the explicit9 approach of detecting RPE use, empirical evidence has been accumulated which concludes that only a minority of firms relies on RPE to compensate executives. Instead, internally based targets which mostly rely on past performance are far more pervasive.10 Notwithstanding that doubts have been raised as to

8

We follow Murphy (2001) in that we assume that the economic incentives arising from negotiation based target setting are similar to using past performance to set targets. Just as target setting based on past performance, negotiations are likely to rely on past performance relative to target as the most important information criterion.
In order to avoid future target increases, managers are likely to withhold current effort in order to be able to exploit the manipulated signal to argue in favor of lower targets in participative budgeting settings (Anderson et al. 2010).
9
While the explicit approach to detecting RPE requires examining information provided by the subject (e.g., by firms in public disclosures), the implicit approach relies on regressing executive compensation on peer performance to obtain estimates of incentive weights of peer performance in managerial compensation.
10
In a survey sample of 177 large US corporations from the year 1997, Murphy (2001) finds that 11% of firms rely on peer performance and another 12% use a mixture of peer and internal information sources as performance targets in their annual incentive plan. Bannister and Newman (2003) examine compensation committee reports of 160 Fortune 250 firms from the year 1992 and identified 28% of them as using RPE in annual and long-term incentive plans. Gong et al. (2011) identify 25.4% of firms in their sample of 1,419 S&P
1500 firms to use RPE in cash and equity based incentive plans based on analysis of fiscal year 2006 proxy

10

whether the explicit approach might underestimate actual RPE use (Black et al. 2011), a significant share of firms uses past information to set performance targets and thereby foregoes the benefits of RPE such as insurance provision against common shocks affecting performance evaluation and strengthened ex ante incentives of effort.
One important drawback of RPE use in contracting is that it practically requires waiting until the end of the period before the target measure peer performance becomes available to evaluate managerial performance. At the same time, empirical evidence reports that the use of budget targets (i.e., targets which are being set ex ante in the beginning of the period) has been pervasive (Umapathy 1987; Eckholm and Wallin 2000; Libby and Lindsay 2010). This finding suggests that firms associate important benefits with having targets in place in the beginning of the period. Even though budgets have been subject to criticism (e.g., Hansen et al. 2003), a variety of reasons-to-budget besides mere performance evaluation and compensation purposes have been put forward. First, budgets may be useful for resource allocation. For instance, using budgeting for both performance evaluation and resource allocation purposes has been found to generate synergies (Fisher et al. 2002). Next, a participative budgeting process may allow managers to convey local information (Shields and
Shields 1998) which otherwise would not be available for decision makers. Furthermore, budgets have been found to be effective for purposes of operational planning, communication of goals and strategy formulation as well as to signal compliance with social norms (Hansen and van der Stede 2004; Covaleski et al. 2007). Overall, this evidence suggests that firms and/or supervisors for a variety of reasons appear to strongly prefer having budget targets in place in the beginning of the period, which precludes the use of RPE ex post and its associated benefits. With this study, we seek to shed light on an area previously neglected by the literature which is whether supervisors who are constrained to budget performance targets in the beginning of the period will incorporate also signals of past peer performance into targets. We argue that weighting past peer performance alongside with past individual performance in setting budget targets to evaluate managers is associated with important benefits. As indicated above, relying on past individual performance as targets deteriorates ex ante incentives to exercise effort. Therefore, contract theory suggests that a more tempered use of past performance information instead of fully exploiting new information signals might

statements. In contrast, Murphy (2001) reports that 77% of firms in her sample rely on strictly internal targets, that is mostly budgets, prior year performance, or both, for the use in annual incentive plans.

11

be a superior strategy in a multi-period setting (Laffont and Tirole 1993, 376). Therefore, commitment to existing contracts might be able to reduce the ratchet effect of withholding effort that results from a strong emphasis of past performance on target setting. Studies on diverse samples of firms find supporting evidence of a moderate use of past performance in target setting. Indjejikian and Nanda (2002) find that executive achievement of target bonus levels is serially correlated. In other words, managers are more likely to earn their target bonus in the current period if they have already achieved their prior period target bonus levels. Indjejikian and Matejka (2006) and Indjejikian et al. (2011) provide further evidence that past individual performance is not fully incorporated into targets. Especially the latter study suggests that unlike badly performing firms, well-performing firms do not revise targets upwards if managers meet or exceed their target, thereby effectively committing to existing levels of managerial compensation for future periods.
We extend this line of research and predict that supervisors will weight past peer performance as a performance signal to set targets. By assigning weight in target setting to past peer performance instead of past individual performance, targets are revised based on common shocks yet less sensitive to managerial effort levels, thereby strengthening commitment to existing contractual arrangements that ensure that well-performing managers continue to earn economic rents. Effectively, this means that in a good economic environment, the target of a manager who exerted high effort and outperformed her peer group is going to be bumped up by a lesser extent than a manager who only benefitted from a favorable economy without exerting much effort. In turn, for adverse economic conditions, managers who failed to meet their target, but still outperform their peers due to high effort levels, will enjoy stronger downward revisions of targets in the future. In both scenarios, incorporating past peer performance into targets ensures that managers who exert high effort earn larger economic rents in future periods by giving them higher targets instead of punishing managers for excess performance relative to target by raising their future target levels. Thus, we hypothesize that targets on a given performance measure do not only reflect past own performance, but also past peer performance.
H1a: Performance targets on a given performance measure increase in past own performance on that measure.
H1b: Performance targets on a given performance measure increase in past peer performance on that measure.

12

Prior studies examine determinants of target setting in a single signal setting (i.e., targets set on past earnings or sales). However, the informativeness principle predicts that manager’s compensation contracts will be rich in performance measures (Holmstrom 1979), given that each performance measure allows supervisors to more accurately infer about managerial performance. If performance measure are aggregated into a single compound, theory predicts that the value of an additional performance measure added to the compound will depend on two informational properties that increase the informativeness of performance evaluation regarding managerial performance (Banker and Datar 1989; Feltham and Xie
1994). First, performance measures which are highly sensitive to a manager’s effort should be weighted more heavily in performance contracts. Second, performance measures which are very noisy (i.e., mostly driven by exogenous shocks) should be weighted to a lesser extent.
Consistent with the prediction of the informativeness principle, prior studies find that accounting performance measures are supplemented by measures of stock market performance (Lambert and Larcker 1987), non-financial performance measures (Ittner et al.
1997; Campbell 2008) or individual performance measures (Bushman et al. 1996) to form a portfolio of performance signals based on which managers are evaluated. These findings suggest that compensation contracts rely on a multitude of performance signals to reward managers. Drawing on the informativeness principle, we argue that signals from various performance measures are incorporated into targets. Recall that we hypothesized under H1b that target revisions will incorporate peer performance in order to ensure higher future rents for managers supplying high levels of current effort. Using multiple instead of a single signal of how managers performed relative to their peers will allow supervisors to more accurately estimate managerial performance on several performance measures, thereby strengthening commitment to ensure higher rents for managers supplying high levels of effort. Therefore, we predict that supervisors will find it useful to not only incorporate peer performance on a given measure, but also peer performance on other measures into performance targets.
H2: Performance targets on a given performance measure increase in past peer performance on other available performance measures.
Our first two hypotheses predict that supervisors will use past peer performance and not solely past individual performance to set performance targets. Prior literature on the use of
RPE suggests that the benefits of RPE increase in the peer group’s capacity to filter common shocks from the manager’s performance measure. Therefore, the performance gains resulting

13

from RPE-related improvements in contract efficiency (i.e., improved risk sharing) are amplified by the extent of common risk shared by the peer group (Matsumura and Shin 2006).
As the benefits of RPE increase in common risk borne by the peer group, the decision to use
RPE is going to depend on whether managerial performance is affected by the same shocks as their peers’ (Gong et al. 2011). Therefore, an RPE scheme is likely to be based on the performance of a peer group which is composed of, for instance, managers that share a large extent of common risk with their respective peers (Albuquerque 2009; Gong et al. 2011).
Furthermore, not only will the composition of the peer group be chosen in a way that provides as much protection against common shocks as possible, but so will be the weights assigned in the course of peer group composition to each individual peer group member
(Dikolli et al. 2012; Black et al. 2011).
The relative weight at which a performance measure is aggregated depends on its informational properties. Banker and Datar (1989) established the notion that two performance measures will be weighted based on the ratio of their respective “signal-tonoise” ratios. Empirical evidence tends to support the conjecture that relative weights used in compensation contracts reflect the informational properties of performance measures. Noise in financial performance measures decreases weight of the latter relative to non-financial performance measures (Ittner et al. 1997). Noise of accounting versus stock based measures of performance will decrease the ratio of weights of one measure versus the other (Lambert and Larcker 1987). 11
Recall that we argue above that RPE is used for target setting to filter noise from performance measures to more accurately infer about managerial effort and incorporate this information in a way such that hard working managers are rewarded with higher future rents resulting from lower targets. The benefits of weighting one peer performance measure versus the other will largely depend on the benefits (i.e., the capacity to filter common shocks) of one measure versus another. Thus, the weights assigned to past peer performance in target setting are likely going to differ by performance measure. First, two different performance measures might be subject to different levels of common shocks, possible affecting the returns from using RPE. Second, peer group composition might be imperfect as peer group construction reflects a trade-off between identifying a portfolio of peers that maximizes common risk and
11

Intuitively, the idea that two measures should be weighted based on their informational properties applies to
RPE as a special case with individual and peer performance as the two signals of performance, as discussed by
Lambert (2001, 24): peer performance will be weighted more heavily relative to individual performance at the extent the two measures are affected by a common shock.

14

the cost of constructing a peer group. Therefore, peer performance based on a composed peer portfolio is a noisy proxy for common shocks affecting performance on a given measure.
Third, supervisors and managers might have little discretion over peer group composition which might be imposed on them (e.g., by top management or consultants). Adjusting weights on peer performance in negotiations might therefore be an effective means for them to restore an optimal weighting of performance measures to insulate aggregate performance based on which managers are evaluated from common risk. Hence, we argue that in a multi-signal setting, supervisors will place more weight on those measures of peer performance where the peer group is highly effective in filtering out common shocks relative to measures where it might be less effective at doing so.
H3: When determining performance targets, the relative weight placed on a given measure of past peer performance increases if peer group capacity to filter common shocks from this performance measure increases from low to high.

RESEARCH SETTING
The research setting for our study is a national employment agency, hereafter referred to as “Employment Agency”. It is a large public entity which is largely self-governing, fully equipped with decision rights and supervised by the country’s Department of Labor.
Employment Agency is publicly financed through taxes and contribution payments to statutory unemployment insurance which is part of the national social security system. Its mission is to provide training, placement services, unemployment benefits and welfare payments to the employment-seeking population as well as act as a broker between employers and employees. It employs well above 100,000 employees and has an annual budget at its disposal which amounts to a double digit billion dollar equivalent.
The organization of Employment Agency is divided into 10 regions, each headed by a regional manager who reports directly to the CEO. Each regional manager is directly responsible for those of the 173 local agencies located on her region’s territory. The agencies are each headed by an agency manager. Employment Agency has defined four different business units which differ by the type of customers they serve (e.g., long-term unemployed).
At the local level, 354 so-called employment centers are serving long-term customers who have been unemployed for longer than 12 months. An employment center manager is supervised by an agency manager (see Figure 1).

15

Figure 1
Organizational Chart of Employment Agency

Headquarters

10
Regions

Reg A

173
Agencies

Reg B

Agency A

Agency B

….

…

Study focus

Supervisor

354
Employment
Centers*

EC A

EC B

…

Manager

Target-based contracting

Performance Evaluation
Employment Agency has undergone a Public Management style reform since the turn of the century which introduced a number of management techniques widely used in the private sector such as performance-based pay, career programs and an elaborate performance measurement system. One cornerstone of the reform was a strong emphasis on control of outcomes as opposed to directives as commonly used in bureaucratic organizations (Ouchi
1979, 1980) such as public administration.12 As a result, managerial performance is evaluated based on performance relative to budget targets. In the beginning of the business year, managers (i.e., employment center managers) contract with their respective supervisors at the higher hierarchical level (i.e., agency managers) on performance targets levels regarding their respective performance measure. Performance measures and incentive weights are revised each year based on the strategic priorities of top management. Managerial performance is evaluated based on the weighted aggregate of target achievements on the respective performance measures. Aggregate as well as single target achievement is capped at 120%.
12

Employment Agency uses the terms “agency model” to express a strong emphasis on contracting on performance targets as opposed to bureaucratic forms of control.

16

Based on their target overall target achievement, managers are assigned a grade, which is A for 110% to 120%, B for 100% to 110% and so forth. A subjective component in performance evaluation exists as supervisor have discretion to adjust the grade by one step if the initial grade falls in the lowest or highest tercile, respectively, and they can put forward a reasonable justification for exercising this discretion. The result of the annual performance evaluation provides a compensation incentive. First, it is determining a manager’s promotion prospects as the manager may signal ability that qualifies for higher paid positions. Second, as the first and only public entity in the country, Employment Agency has introduced performance based pay for managers. Depending on the size of their unit, employment center managers can earn up to 20% of their monthly or annual salary as an annual bonus if they reach the highest grade
A, 15% for a grade B etc.
Due to the absence of accounting profits which are commonly used in private entities,
Employment Agency relies on a set of financial and non-financial performance measures to assess performance of the business unit serving long-run unemployed customers (see Table
1). Since the business unit focused on long-run unemployed customers is funded from taxpayer money, it is exposed to strong political pressure to operate as effectively as possible, that is to reduce the duration of average unemployment and the number of customers. The two most important performance measures to assess employment center performance are “labor market integrations per customer” (i.e., the share of employment seeking customers who found employment within the business period of one year) (IQ) and “welfare payments”
(payment amount spent on welfare payments within the business period of one year) (WP).
They are assigned the highest incentive weights and are also the only two measures with nonzero incentive weights over the entire study period.

Table 1
Performance Measures and Weights at Employment Agency Business Unit “Long Run Unemployed Customers”
Performance Measure
Description
Target
2007
2008
2009
negotiated?
Weight
Weight
Weight

2010
Weight

Money Amount of
Welfare Payments

Yes

33,3%

33,3%

35%

40%

Integrations per
Customer

Improve rate of integrations into the labor market

Yes

33,3%

33,3%

15%

20%

Integrations per
Customer U25

Improve rate of integrations of younger labor market participants (below 25 years) into the labor market

Yes

33,3%

33,3%

-

-

Customers with long term dependency

Reduce number of customers with long term unemployment
(>24 months)
Improve customer satisfaction

Yes

-

-

15%

15%

No

-

-

20%

10%

Process quality index

67%
Business related measures Reduce dependence of recipients on welfare payments

Improve process quality

No

-

-

15%

15%

Employee training
Promotions

Conduct employee training
Identify managers eligible for promotion No

Customer satisfaction index 33%
Employee related measures 100%

18

Target Setting Process
Given the interest of this study in examining and modeling how available performance signals are incorporated into performance targets, it is important to understand the target setting process in our specific field setting. The two-stage target setting process at
Employment Agency works as follows. In the first stage, initial target proposals for the local units such as regions, agencies and employment centers are calculated by headquarters according to a formula. The formula is a function of the overall economic outlook for the upcoming year, past individual performance and past peer performance. More specifically, targets are calculated as growth rates relative to prior year levels (e.g., integrations per customer), which effectively leads to a strong emphasis of past performance levels. A peer performance component is built into targets as all managers who are not in the pre-defined quantile (e.g., first quartile) receive a target markup. The target markup increases proportionally in their relative performance shortfall as compared to the peer group. The peer groups used to determine peer performance have been constructed by cluster analysis based on several variables which determine the structure of the local labor market (e.g., unemployment rate, share of long-term unemployed). Prior to that, the structural determinants have been identified by regression analysis based on their ability to explain a large share of variance of the integrations per customer performance measure. Besides their use for target setting, the peer groups are used for monitoring target achievement and performance relative to peers during the year.
The target proposals serve as a basis for the second stage of the target setting process, the subsequent negotiations of targets at the local level. These negotiations are carried out between the employment center manager and her supervisors, that is the agency manager and a representative from the regional management.13 Throughout the negotiations, managers at all levels are supported by financial analysts and an accounting information system which provides unrestricted access to comprehensive performance data on all employment centers.
After the negotiations, targets are submitted by employment center managers, reviewed by the
13

Because their target is the aggregate of all targets of lower level units under their responsibility, altering target proposals of employment centers will also affect the agency manager’s and regional manager’s target. Therefore, supervisors might have incentives to pad employment centers targets. On the other hand, systematic padding across all employment centers is likely to be detected by headquarters now or at a later point (when employment centers are performing poorly relative to their respective peer groups), so supervisors might instead rather prefer to explore how targets can be redistributed across all local units in their area of responsibility. Reallocation of targets might achieve gains from recalibration of compensation risk and fairness perceptions even though it is likely to involve confrontation costs (Bol et al. 2010). Anecdotal evidence from interviews with headquarters’ task force members indicates that supervisor aim to set accurate targets.

19

headquarters and finally approved by the regions, unless renegotiation is deemed necessary by regional or agency managers.

RESEARCH DESIGN
Sample
To study the target setting process at Employment Agency, we collected monthly performance and operational data on 354 employment centers spanning the years 2007 till
2010 from the organization’s accounting information system. Additionally, we collected annual macroeconomic indicators on the regional labor markets served by the employment centers from Employment Agency’s statistics department. From the initial 1,062 observations, we had to drop 88 observations due to missing values on one or more variables or mergers which precluded the reconstruction of data on actual performance of merged units before the merger. Our final sample consists of 974 employment center years. We conducted interviews with managers and employees at headquarters and gathered qualitative data such as planning guidelines, performance reports and templates used for assessment of managerial performance to further our understanding of Employment Agency’s performance evaluation practices.

Dependent Variable
To study which information signals are incorporated into targets, we examine target revisions of employment center targets with respect to Employment Agency’s “integrations per customer” performance measure IQ used in compensation contracts (see section III). This performance measure indicates employment center performance at placing long-run unemployed customers into employment or in supporting them to start their own business.
Due to the two-stage nature of the target setting process outlined under section 0., we distinguish between initial target proposals IQ_ORt14 as calculated and disseminated by headquarters and final targets IQ_TRt resulting from negotiations. Data availability on both
IQ_ORt and IQ_TRt allows us to disentangle the effect of initial target proposals stemming from headquarters and the incremental adjustments resulting from local negotiations between employment center managers and their supervisors, the agency managers, on targets.

14

We label initial target proposal as “IQ_OR” because Employment Agency refers to its initial target proposals as “Orientation Values”.

20

Table 2
Variable Definitions
IQ_ACTt,i

 Actual labor market integrations per customer of employment center i in year t

IQ_TRt,i

 Target labor market integrations per customer

IQ_ORt,i

 Initial target proposal on labor market integrations per customer as calculated by headquarters

IQ_ACTt,i –IQ_TRt,i

 Actual employment center labor market integrations per customer relative to target (difference in levels)

IQ_ACT_RANKt,i

 Peer group rank based on actual labor market integrations per customer (adjusted for peer group size: 0 indicates lowest rank and good relative performance, 1 indicates highest rank and poor relative performance)

WP_ACTt,i

 Actual employment center welfare payments

WP_TRt,i

 Target employment center welfare payments

%WP_ACTt,i

 Actual employment center savings in welfare payments relative to prior year actual welfare payments

%WP_TRt,i

 Target employment center savings in welfare payments relative to prior year’s actual welfare payments

(WP_TRt,i – WP_ACTt,i) /
WP_TRt,i

 Actual employment center welfare payments relative to target (scaled by target)

%WP_ACT_RANKt,i

 Rank based on actual welfare payment savings relative to peer group (adjusted for peer group size: 0 indicates lowest rank and good relative performance, 1 indicates highest rank and poor relative performance)

%URPeer, t,i

 Percentage change in peer group mean unemployment rate
(relative to prior year)

%GDPpCt,i

 Percentage change in gross domestic product per capita in employment center district (relative to prior year)

SIZEt,i

 Logged customer base size

PEER_SIZEi

 Number of peer group members

COMMON_RISK(IQ)i

 Overlap between actual and optimal peer group members with respect to integrations per customer measure (optimal peer group member for each employment center chosen based on correlation coefficient between given employment center performance and any other employment center’s performance on integrations per customer measure for a 53 month time series ranging from January 2007 till May 2011)

COMMON_RISK(WP)i

 Overlap between actual and optimal peer group members with respect to welfare payments (optimal peer group composition follows procedure for COMMON_RISK(IQ)i)

year2009, year2010

 Dummy variable indicating year of observation

21

Explanatory Variables

Past Individual Performance
To examine the weight assigned to different performance signals in target setting, we model the influence of past individual and peer performance on performance targets. Broadly speaking, prior studies have assessed the extent of reliance on past performance in target setting based on the size and significance of the influence of past target achievement, that is past performance relative to target, on performance targets. We measure target achievement as excess performance or performance shortfall relative to target (e.g., Bouwens and Kroos
2011, Leone and Rock 2002). IQ_ACT –IQ_TRt-1 captures excess performance relative to target on IQ whereas (WP_TRt-1 – WP_ACTt-1) / WP_TRt-1 captures target achievement on the second performance measure amount spent on welfare payments WP.15

Past Peer Performance
To investigate whether supervisors not only rely on past individual, but also past peer performance to set performance targets used in contracts, we estimate the impact of past peer performance regarding two performance measures IQ and WP on target IQ_TRt.16 We rank employment centers based on their actual performance IQ_ACTt-1 and measure peer performance as the rank of a given employment center in its peer group after adjusting for peer group size:
IQ _ ACT _ RANK t 1 

Rank IQ _ ACTt 1 
PEER _ SIZE

Likewise, we measure peer performance on the second performance measure WP by ranking all employment centers based on their actual savings %WP_ACTt-1 on welfare payments:

15

Because less welfare payments indicate superior performance, we reverse the sign on the difference between actual and target value. Since there are substantial size differences across employment centers, we scale the difference by the target value to make it comparable across employment centers of different size.
16
With respect to each performance measure, a high rank indicates strong peer performance while a low rank indicates weak peer performance, respectively. Thus, a low rank indicates that an employment center manager has outperformed her peer group while a high rank indicates that she has, in turn, been outperformed by her peers. 22

%WP _ ACT _ RANK t 1 

Rank %WP _ ACT _ RANK t 1 
PEER _ SIZE

WP _ ACTt  2  WP _ ACTt 1 
Rank 

WP _ ACTt  2



PEER _ SIZE

Control Variables
To control for differences in the target setting process, we specify the following control variables. First, we control for the macroeconomic environment which is likely to influence demand and supply in the labor market and thereby affect current expected performance of employment centers. A high local unemployment rate reflects reduced demand for labor in a local economy and affects employment center performance (Cragg 1997; Courty and
Marschke 2004). Thus, we include the average percentage change in unemployment rate at the peer group level %∆URPeer,t-1 to capture the current dynamics in the labor market.
Additionally, local gross domestic product (GDP) per capita is going to reflect an economy’s overall performance (Mankiw 2008). Thus, we include the percentage change of gross domestic product per capita relative to prior year %∆GDPpCt-1 to proxy for managerial expectations regarding the local economic environment.
Second, prior research has indicated that organizational size is likely to affect target setting (Bol et al. 2010). Especially, size is likely to be correlated with organizational complexity, a potential driver of target setting. Due to the lack of availability of data regarding headcount or asset base for a public entity, we measure size based on the logged number of customers served by the employment center (SIZE). Finally, we include year indicators to capture year effects reflecting the general condition of the national economy.

Peer Group Capacity to Filter Common Risk
Hypothesis H3 predicts the relative weight associated with two different measures of peer performance to depend on the capacity of the peer group to filter out common shocks from one performance measure relative to the other. Contract theory defines common risk as the correlation between individual and peer performance (e.g., Lambert 2001). Accordingly, a high correlation between individual and peer performance is considered a consequence of exposure to common shocks which explains a substantial share of individual performance variance. Therefore, those managers for which the performance of the available peer group

23

exhibits a high degree of correlation with their own performance are more likely to benefit from the use of RPE (Matsumura and Shin 2006; Gong et al. 2011).
To construct a measure for common risk and evaluate a peer group’s capacity to bear common risk, we do the following. First, we construct an optimal portfolio of peers from all employment centers. For each of the employment centers, we collect a 53 month time series ranging from January 2007 till May 2011 of each of the two performance measures (IQ_ACT and WP_ACT). To proxy for the extent of exposure to common shocks of a given employment center with any other, we compute pairwise correlations between performance of a given employment center with each of the other remaining employment centers, where performance is measured by one month differences in IQ_ACT and WP_ACT, respectively. Having ranked employment centers based on time series correlations on each of the two performance measures, we replace actual peer group members by those employment centers with the highest time series correlations to construct optimal peer groups of same group size as the actual peer group. Thus, we choose the optimal peer groups such that the portfolio of peers maximizes the average correlation in performance between a given employment center and its peer group members.
Second, we evaluate peer group quality (i.e., its ability to insulate a given employment center from common risk) based on the overlap between actual and optimal peer group members. Specifically, we measure the extent of common risk bearing ability by
COMMON_RISK(IQ) and COMMON_RISK(WP) as the share of overlap between actual and optimal peer group composition with respect to either of the two performance measures
IQ_ACT and WP_ACT, respectively. Subsequently, we split the sample by half into high and low levels of common risk bearing ability on each of the two performance measures.

EMPIRICAL MODELS AND RESULTS
Descriptive Statistics and Univariate Correlations
Table 3 provides descriptive statistics on all variables used in the analysis, while Table
4 provides pairwise correlations coefficients among all variables. Employment center performance against target on both performance measures, IQ_ACTt-1 – IQ_TRt-1 and
(WP_TRt-1 – WP_ACTt-1) / WP_TRt-1, shows substantial cross-sectional and time series variance. Both above-mentioned measures of target performance are positively related to

24

initial target proposals IQ_ORt and final targets IQ_TRt, which is consistent with a strong emphasis on past performance for target setting. As expected, both measures of employment center performance IQ_ACTt-1 and %WP_ACTt-1 are strongly correlated, most likely due to exposure to the same shocks, managerial effort and ability. The two variables capturing the macroeconomic environment, %∆URPeer,t-1 and %∆GDPpCt-1 are highly correlated with target achievement on both performance measures, IQ_ACTt-1 – IQ_TRt-1 and (WP_TRt-1 –
WP_ACTt-1) / WP_TR5–1, which suggests that performance evaluation is subject to a significant amount of noise resulting from exogenous shocks. Finally, the two measures of peer group capacity to bear common risk, COMMON_RISK(IQ) and COMMON_RISK(WP), are strongly, but not perfectly correlated. This indicates that the capacity of a given peer group to insulate managers from common shocks may differ by performance measure.

Table 3
Descriptive Statistics
2008
Variable

N

Mean

Std.

2009
Min.

Max.

N

Mean

Std.

2010
Min.

Max.

N

Mean

Std.

Min.

Max.

IQ_TRt

322

24.61

3.97

15.2

37.89

326

22.83

3.12

14.5

34.36

326

16.52

2.51

10.19

26.38

IQ_ORt

322

25.63

4.03

16.65

39.13

326

24.3

3.46

15.83

36.03

326

16.41

2.39

10.29

24.78

IQ_TRt-1

322

21.65

3.87

13.47

34.3

326

24.6

3.97

15.2

37.89

326

22.83

3.12

14.5

34.36

IQ_ACTt-1

322

23.07

4.08

13.54

36.57

326

23.32

3.51

14.45

35.44

326

19.19

2.74

12.17

30.15

IQ_ACTt-1 – IQ_TRt-1

322

1.43

2.37

-5.23

10.84

326

-1.28

1.79

-7.34

4.49

326

-3.64

1.8

-9.5

1.06

IQ_ACT_RANKt-1

322

0.53

0.29

0.02

1

326

0.52

0.29

0.02

1

326

0.53

0.29

0.02

1

WP_ACTt-1(in millions)

322

38.43

49.37

3.51 473.04

326

36.52

47.83

2.9 457.79

326

37.35

48.72

3.12 463.57

WP_TRt-1(in millions)

322

39.15

48.69

4.38 470.49

326

36.02

46.96

46.75

2.68 444.53

%WP_ACTt-1

322

7.97

4.74

-1.24

24.4

326

%WP_TRt-1

322

3.54

1

0

8.4

(WP_TRt-1 – WP_ACTt-1) / WP_TRt-1

322

4.59

4.88

-5.46

%WP_ACT_RANKt-1

322

0.48

0.29

%URPeer, t-1

322

-0.19

%GDPpCt-1

322

3.16

446.2

326

35.44

7.16

3.93 -11.74

21.33

326

-3.03

326

8.35

3.03

3.04

17.5

326

3.79

21.57

326

-1.2

3.59 -15.26

10.8

326

-7.16

0

0.98

326

0.5

0.29

0

0.98

326

0.04

-0.24

-0.1

326

-0.16

0.04

-0.23

-0.09

0.04

0.03

-0.04

0.18

326

0.02

0.04

-0.19

0.21

SIZE(unlogged, in 1,000’s)t-1

322 101.62

90.11

326 101.19

89.47

PEER_SIZEi

322

326

10.91

COMMON_RISK(IQ)i

322

0.19

0.12

0

0.52

326

0.19

0.12

0

0.52

326

0.19

0.12

0

0.52

COMMON_RISK(WP)i

322

0.21

0.14

0

0.55

326

0.21

0.14

0

0.55

326

0.21

0.14

0

0.55

34.6

10.9

17.9 891.94
13

50

34.48

17.89 899.51
13

50

4.29 -17.04

7.6

1.46

1

12

5.07 -23.71

3.15

0.49

0.29

0

0.98

326

0.1

0.09

-0.05

0.24

326

-0.03

0.03

-0.17

0.04

326 101.84

90.69

326

10.91

34.48

17.9 916.04
13

50

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16

IQ_TRt
IQ_ORt
IQ_TRt-1
IQ_ACTt-1
IQ_ACTt-1 – IQ_TRt-1
IQ_ACT_RANKt-1
%WP_ACTt-1
%WP_TRt-1
(WP_TRt-1 – WP_ACTt-1) / WP_TRt-1
%WP_ACT_RANKt-1
%URPeer, t-1
%GDPpCt-1
SIZEt-1
PEER_SIZEi
COMMON_RISK(IQ)i
COMMON_RISK(WP)i

1
1
0.98
0.5
0.92
0.59
-0.43
0.77
0.29
0.66
-0.13
-0.75
0.51
-0.22
0.2
-0.2
-0.15

2

3

1
0.49
0.91
0.59
-0.39
0.8
0.35
0.67
-0.13
-0.8
0.52
-0.19
0.2
-0.19
-0.13

1
0.73
-0.34
-0.46
0.26
0.51
0.03
-0.21
-0.06
-0.07
-0.3
0.36
-0.29
-0.16

Table 4
Correlation Table
4
5
6
7

1
0.4
-0.61
0.63
0.39
0.48
-0.22
-0.52
0.3
-0.26
0.27
-0.26
-0.17

1
-0.23
0.52
-0.16
0.62
-0.03
-0.63
0.5
0.05
-0.12
0.03
-0.02

1
-0.17
-0.03
-0.17
0.35
-0
-0.01
0.13
-0.05
0.03
-0.04

1
0.33
0.89
-0.46
-0.84
0.58
-0.15
0.07
-0.07
-0.04

8

9

10

11

1
-0.14
-0.04
-0.3
0.09
-0.09
0.22
-0.14
-0.06

1
-0.46
-0.74
0.58
-0.12
-0.04
0.01
-0.01

1
-0
-0.09
0.15
0
0.02
-0.04

1
-0.66
0.05
0.01
0.03
0.03

12

13

14

15 16

1
-0.02 1
-0.02 0.02 1
0.06 0.27 0.09 1
0.01 0.31 0.33 0.5 1

27

Determinants of Initial Target Proposals from Headquarters
The purpose of this study is to investigate which signals of past performance are incorporated into performance targets. Due to the two-stage target setting process at
Employment Agency, understanding how much weight is placed on the respective performance measures requires to study (1) how headquarters sets initial target proposals and
(2) how initial target proposals are adjusted in negotiations to reflect an optimal weighting of various performance signals.
Planning guidelines and interviews with managers from headquarters reveal that past performance IQ_ACTt-1 and peer performance IQ_ACT_RANKt-1 are used to calculate initial targets IQ_ORt. Each year’s target proposals are calculated based on a formula which has been determined by top management. First, past performance IQ_ACTt-1 is expected to grow by a certain percentage a given year, thereby putting a strong emphasis on past performance.
Second, employment centers which rank below a predefined top quantile on IQ_ACTt-1 are deemed to be operating inefficiently and receive a target mark-up which grows proportionally with the performance shortfall relative to efficient peer group members.
To verify the impact of both past peer and individual performance on target proposals, we estimate the following equation

IQ _ ORt   0  1 * IQ _ ACTt 1   2 *IQ _ ACT _ RANKt 1   3 * year2009
  4 * year2010   IQ _ OR

(1)

by OLS. We address potential concerns regarding bias in OLS standard errors arising from heteroskedasticity and serial correlations by using robust standard errors clustered by employment center (Rogers 1993).
Table 5, column 1 presents the results from estimation of equation (1). Consistent with qualitative evidence, the positive and highly significant effect of IQ_ACTt-1 (α1 = 1.030, t =
165.68) on IQ_ORt suggests that headquarters relies heavily on past performance to set initial target proposals. Likewise, the positive highly significant effect of IQ_ACT_RANKt-1 (α2 =
1.551, t = 18.54) implies that employment centers which were outperformed by their peers receive higher targets than if they outperformed their peer group. The results remain

28

qualitatively unchanged when we include IQ_TRt-1 (see column 2). Thus, relative to past individual performance, past targets provide little incremental explanatory power for target proposals. Based on the large t-statistics and the high R2 of 98%, we conclude that we managed to confirm qualitative assertions regarding the use of past individual and peer performance by empirically replicating Employment Agency’s formulaic approach to determine target proposals. Our findings provide single firm evidence that top management incorporates past peer performance alongside past own performance to ensure that wellperforming managers are able to earn future rents by lowering their targets.

Table 5
Determinants of Initial Target Proposals
OLS
(1) coef (t-Stat)
IQ_ORt

Dependent variable:
IQ_ACTt-1

α1

IQ_ACT_RANKt-1

α2

year2009

α3

year2010

α4

IQ_TRt-1

OLS
(2)
coef
(t-Stat)
IQ_ORt

1.030
(165.68)***
1.551
(18.54)***
-1.589
(40.45)***
-5.226
(69.73)***

α5

_cons
R2
N

1.049
(5.92)***
0.98
974

1.102
(80.93)***
1.729
(20.00)***
-1.401
(27.21)***
-4.865
(47.83)***
-0.069
(6.25)***
0.797
(4.19)***
0.99
974

*,** and *** indicate significance at less than the 10 percent, 5 percent and 1% level, respectively, based on two-tailed t-tests
Rogers (1993) standard errors clustered by employment center

Determinants of Target Adjustments in Negotiations
Next, we turn to the main focus of our paper which is to investigate which type of performance signals are incorporated in negotiations between agency managers and employment center managers to adjust initial target proposals. By specifying IQ_TRt as a

29

dependent variable and IQ_ORt as an independent variable17, all effects are to be interpreted as incremental effects on final targets arising from negotiations. Furthermore, we include past individual performance against target and peer performance on IQ and WP, respectively, as well as controls for the macroeconomic environment18, size and years:

IQ _ TR t   0   1 * IQ _ OR t   2 * ( IQ _ ACT t 1  IQ _ TR t 1 )
  3 * IQ _ ACT _ RANK t 1   4 * year 2009   5 * year 2010
  6 * % AQ Peer ,t 1   7 * % GDPpC t 1   8 * SIZE t 1
  9 * (WP _ TR t 1  WP _ ACT t 1 ) / WP _ TR t 1
  10 * % WP _ ACT _ RANK t 1   IQ _ TR

(2)

We simultaneously estimate a system of structural equations (1) and (2) without imposing any constraints on the error terms. This is similar to seemingly unrelated regression (SUR)
(Zellner 1962) in that it exploits correlations between error terms to improve efficiency of estimation.19 The results of the estimation which are presented in Table 6, column 2 reveal the following. IQ_ACTt-1 – IQ_TRt-1 is not significantly related to IQ_TRt (β2 = 0.019, z = 1.44) based on a two-tailed test. We conclude that there appears to be no further target adjustment for past own performance beyond the positive weight placed on past own performance through headquarters. Overall, targets seem to reflect past own performance, as hypothesized under H1a. Next, IQ_ACT_RANKt-1 appears to be negatively related to target adjustments in the course of negotiations (β3 = -0.698, z = 5.19) at the 1% significance level. This finding implies that employment centers which are outperformed by their peers receive smaller adjustments of target proposals compared to employment centers which have outperformed their peers. Thus, contrary to the intuition suggested under H1b, supervisors appear to adjust targets such that poor relative performers’ target adjustments are smaller than high relative performers’. This effectively results in a negative weight placed on peer performance
IQ_ACT_RANKt-1 during target adjustments. Recall that the overall weight placed on performance signals reflected in targets depends on the joint effect of headquarters’ target proposals and subsequent negotiations. Consistent with H1b, the overall weight placed on
17

Equivalently, we could regress the difference IQ_TRt – IQ_ORt on the remaining explanatory variables.
However, the approach we chose has the advantage of not constraining the coefficient on IQ_ORt to 1.
18
We also specified lagging and leading indicators of the macroeconomic control variables for robustness checks. 19
We estimate the system of equations using STATA’s sem command which, unlike the SUR estimator, allows to adjust standard errors for heteroskedasticity and serial correlation.

30

IQ_ACT_RANKt-1 is significantly positive (α2*β1 + β3 = 1.551 * 0.9 – 0.698 = 0.697, z = 4.72) at the 1% significance level. We interpret this as evidence that initial target proposals calculated by headquarters appear to place too much weight on past peer performance on IQ.
To restore optimal weighting, negotiations partly reverse the weight placed on
IQ_ACT_RANKt-1 by headquarters.20
Next, the results with respect to H2 suggest that supervisors incorporate information from performance signals other than IQ into targets. Consistent with H2, target adjustments regarding IQ are positively associated with peer performance %WP_ACT_RANKt (β10 =
0.868, z = 7.55) at the 1% significance level. As higher ranks represent poor relative performance, employment center managers who outperform their peers based on their realized savings %WP_ACTt-1 will receive lower targets IQ_TRt than their peers with inferior relative performance on that measure. Recall from above that %WP_ACT_RANKt-1 is not used by headquarters to calculate initial target proposals. This finding suggests that supervisors will incorporate signals from others sources that were not considered by headquarters to determine initial targets in order to ensure that well-performing employment center managers receive lower targets in future periods. Furthermore, the positive effect of target achievement
(WP_TRt – WP_ACTt) / WP_TRt on the measure WP (β9 = 0.041, z = 4.59) suggests that supervisors might also factor in past individual performance on other performance measures when setting targets IQ_TRt. Overall, we present empirical evidence from a setting in which multiple signals of performance are provided that suggests that supervisors will use multiple signals of peer performance to set targets. In doing so, they reward managers who consistently outperform their peers on multiple measures with lower targets while current period inferior relative performance on multiple measures is punished by higher targets.

20

The above-mentioned negative weight placed on IQ_ACT_RANKt-1 during target adjustments might be simply due to supervisors’ attempt to spare poor relative performers who are subject to severe target increases imposed on them by headquarters. To address this concern, we run the following robustness check. We split the sample at median relative performance IQ_ACT_RANKt-1 into high and poor relative performers. If the conjecture was true and the effect was entirely driven by supervisor attempts to spare poor relative performers, we should not be able to find that target adjustments place a negative weight on IQ_ACT_RANKt-1 for the high relative performance subsample. However, the lack of statistical difference of the significantly negative response coefficients β3 between the two subgroups indicates the opposite.

Table 6
Determinants of Final Targets: System of Equations Estimation of RPE Use in Target Setting
SUR
(1) coef (z-Stat)

IQ_ORt

β1

Performance measure “Integrations per customer” IQ:
IQ_ACTt-1 – IQ_TRt-1 β2 IQ_ACT_RANKt-1
Control variables: year2009 β3

β4

year2010

β5

%URPeer,t-1

β6

%GDPpCt-1

β7

SIZE t-1

β8

_cons

SUR
(3)
coef
(z-Stat)

SUR
(4)
coef
(z-Stat)

IQ_TRt

Dependent variable:

SUR
(2)
coef
(z-Stat)

System OLS
(5)
coef
(z-Stat)

IQ_TRt

IQ_TRt

IQ_TRt

IQ_TRt

0.905
(67.28)***

0.900
(68.77)***

0.900
(67.85)***

0.902
(68.13)***

0.906
(68.46)***

0.034
(2.60)***
-0.488
(3.70)***

0.019
(1.44)
-0.698
(5.19)***

0.018
(1.36)
-0.868
(5.29)***

0.016
(1.26)
-0.696
(3.96)***

0.007
(0.51)
-0.656
(3.69)***

-0.375
(5.22)***
1.151
(6.44)***
-1.967
(3.88)***
2.239
(3.54)***
-0.191
(4.30)***
3.326
(4.62)***

-0.220
(3.06)***
1.231
(6.88)***
-0.921
(1.66)*
2.471
(3.89)***
-0.200
(4.74)***
3.264
(4.71)***

-0.212
(2.94)***
1.228
(6.85)***
-0.855
(1.54)
2.501
(3.96)***
-0.204
(4.53)***
3.321
(4.93)***

-0.212
(2.94)***
1.228
(6.87)***
-0.818
(1.46)
2.385
(3.78)***
-0.200
(4.44)***
3.125
(4.66)***

-0.374
(5.32)***
0.163
(1.34)
2.521
(5.57)***
2.172
(3.38)***
-0.224
(4.79)***
4.080
(6.05)***

“Welfare Payments” WP:
(WP_TRt-1 – WP_ACTt-1) / WP_TRt-1
%WP_ACT_RANKt-1
Interaction effects:
COMMON_RISK_HIGH(IQ)

β10

β12

COMMON_RISK_HIGH(IQ)*
IQ_ACT_RANKt-1
COMMON_RISK_HIGH(IQ)*
%WP_ACT_RANKt-1
COMMON_RISK_HIGH(WP)*
IQ_ACT_RANKt-1
COMMON_RISK_HIGH(WP)*
%WP_ACT_RANKt-1

0.043
(4.86)***
1.032
(7.06)***
-0.012
(0.08)
-0.035
(0.58)
0.355
(1.82)*
-0.296
(1.67)*

β11

COMMON_RISK_HIGH(WP)

N

0.041
(4.59)***
0.868
(7.55)***

β9

β13 β14 β15 β16 974

974

974

*,** and *** indicate significance at less than the 10 percent, 5 percent and 1% level, respectively, based on two-tailed z-tests
Rogers (1993) standard errors clustered by employment center

0.043
(4.82)***
1.027
(6.37)***

0.035
(3.99)***
0.826
(5.14)***

-0.146
(1.05)
0.287
(2.07)**
0.609
(3.05)***
-0.250
(1.36)
-0.591
(2.93)***
-0.053
(0.29)

-0.188
(1.41)
0.230
(1.73)*
0.584
(2.85)***
-0.251
(1.41)
-0.540
(2.62)***
0.011
(0.06)

974

974

Table 7
Subgroup Analysis of Marginal Effects
Sample Split by
COMMON_RISK(IQ)
Low
High
(1)
(2)
Marginal effect of peer performance integrations per customer on target
IQ _ TRt
  2 1   3,subgroup
IQ _ ACT _ RANK t 1
Marginal effect of peer performance in welfare payment savings on target
IQ _ TRt
 10 ,subgroup
 % WP _ ACT _ RANK t 1
Ratio of marginal effects
 2 1   3,subgroup

10,subgroup
Test for differences in ratios across high and low conditions of common risk21
( 2  1   3,high ) *  10 ,low
= ( 2  1   3,low ) *  10 , high

COMMON_RISK(WP)
Low
High
(3)
(4)

0.567
(z = 3.48)***

0.930
(z = 5.74)***

0.844
(z = 4.94)***

0.506
(z = 2.92)***

1.034
(z = 6.83)***

0.709
(z = 5.49)***

0.911
(z = 6.10)***

0.794
(z = 5.94)***

0.549
(z = 2.84)***

1.311
(z = 3.60)***

0.927
(z = 3.47)***

0.638
(z = 2.36)**

χ2= 5.19**, p = 0.0227

χ2= 0.83, p = 0.3626

Equation (1) and (3) simultaneously estimated with no constraints imposed on error correlations between 1st and 2nd stage equations. No constraints imposed across subgroups with regard to error variances, error covariances and above-mentioned coefficient estimates. Rogers (1993) standard errors clustered by employment center.

21

We choose a multiplicative specification of the test which is considered more reliable than a ratio specification (Phillips and Park 1988).

34

Finally, to investigate whether relative weights assigned to the two measures of peer performance, IQ_ACT_RANKt-1 and %WP_ACT_RANKt-1, vary based on the capacity of the peer group to filter out common shocks as hypothesized under H3, we estimate a system of equations (1) and (3). Equation (3) allows the weights placed on both measures of peer performance in negotiations to vary by low and high ability of the peer group to filter out common shocks on the respective measures. More specifically, we interact both
IQ_ACT_RANKt-1 and %WP_ACT_RANKt-1 with a dummy variable indicating high ability to bear common

risk

on

IQ

(COMMON_RISK_HIGH(IQ))

and

WP

(COMMON_RISK_HIGH(WP)), respectively:

IQ _ TRt   0  1 * IQ _ ORt   2 * ( IQ _ ACTt 1  IQ _ TRt 1 )
  3 * IQ _ ACT _ RANK t 1   4 * year 2009   5 * year 2010
  6 * %AQPeer ,t 1   7 * %GDPpCt 1   8 * SIZEt 1
  9 * (WP _ TRt 1  WP _ ACTt 1 ) / WP _ TRt 1  10 * %WP _ ACT _ RANK t 1
 11 * COMMON _ RISK _ HIGH ( IQ)
 12 * COMMON _ RISK _ HIGH (WP)
 13 * COMMON _ RISK _ HIGH ( IQ) * IQ _ ACT _ RANK t 1
 14 * COMMON _ RISK _ HIGH ( IQ) * %WP _ ACT _ RANK t 1
 15 * COMMON _ RISK _ HIGH (WP) * IQ _ ACT _ RANK t 1
 16 * COMMON _ RISK _ HIGH (WP) * %WP _ ACT _ RANK t 1   IQ _ TR

(3)

Results are presented in Table 6, column 3, and reveal the following. If the capacity of the peer group to bear common risk with respect to performance measure IQ is high, the weight placed on peer performance on that measure for target setting purposes increases significantly
(column 3, β13 = 0.355, z = 1.82) as compared to lower levels of risk bearing ability of the peer group for the same measure. Conversely, the weight placed on peer performance with respect to WP in setting target IQ_TRt decreases significantly (column 3, β14 = 0.296, z =
1.67) if common risk borne by the peer group with respect to IQ increases from low to high levels. This suggests that the relative weight placed on peer performance IQ_ACT_RANKt-1 vis-à-vis %WP_ACT_RANKt-1 increases in the extent that the peer group is filtering common shocks affecting IQ_ACTt-1. Consistent with H3, this finding suggests that for those peer groups that are highly effective in filtering common risk from IQ, negotiations will lead to a shift of incentive weight from peer performance %WP_ACTt-1 to IQ_ACT_RANKt-1.

35

Similarly, results from Table 6, column 4 indicate the following: when peer group capacity to filter out common shocks from %WP_ACTt-1 is high, the weight placed on relative performance IQ_ACT_RANKt-1 is significantly lower (β15 = -0.591, z = 2.93) than if capacity to bear common risk with respect to WP is low. We do not find a significant change in the absolute weight on %WP_ACT_RANKt-1 if the peer group’s common risk bearing ability with respect to WP increases from low to high (β16 = -0.053, z = 0.29). Nevertheless, our findings still support H3 which predicts a decrease in the ratio of weights placed on the two measures of peer performance, IQ_ACT_RANKt-1 vis-à-vis %WP_ACT_RANKt-1 if the capacity to bear common risk with respect to WP increases from low to high levels.22
In order to facilitate interpretation of results, we estimated a system of equations (1) and
(2) while allowing coefficients on IQ_ACT_RANKt-1 and %WP_ACT_RANKt-1 to vary across low and high levels of common risk with regard to IQ and WP, respectively. We calculate the relative weights, that is the ratio of IQ_ACT_RANKt-1 over %WP_ACT_RANKt-1 for high and low levels of common uncertainty with respect to either IQ or WP. The estimates are reported in Table 7. Consistent with the results presented in Table 6, the ratio increases (decreases) if the ability of the peer group to filter common risk from IQ_ACTt-1 (%WP_ACTt-1) increases from low to high levels.23
We interpret this as evidence that relatively more weight is placed in target setting on measures of peer performance that are less noisy in the sense that they are more effective at absorbing a larger amount of common risk. By placing a larger weight on the measure of peer performance that is subject to a high level of common risk relative to a second measure that is less so, supervisors can more reliably distinguish between well and badly performing managers by filtering out common shocks. This allows them to more reliably reward well performing managers with higher future rents stemming from lower targets.

22

We carried out robustness checks to verify whether our results are sensitive to outliers. Our results are robust to winsorizing of continuous variables at 1% and 5% at each tail. Furthermore Cook’s distance failed to identify any outliers at conventional levels (Cook’s D > 1).
23
We test for differences in ratios based on a χ2 test. While we find a significant difference in relative weight assigned to the two performance measures across low and high levels of peer group risk bearing ability with respect to IQ (χ2= 5.19, p = 0.0227), we fail to detect any difference in ratio across low and high levels of peer group risk bearing ability with respect to WP (χ2= 0.83, p = 0.3626).

36

CONCLUSION
Using target and performance data on 354 employment center units of a national employment agency, we investigate the use of available peer performance as a signal to set budget targets in a specific field setting where managerial performance is evaluated only against targets budgeted in the beginning of the year. Our field setting allows us to observe how discretion to adjust initial target proposals provided by headquarters is exercised at the local level and which signals that are available from the accounting information system are incorporated when making these target adjustments. We provide case study type evidence of how headquarters incorporate peer performance into targets by replicating the agency’s target setting procedure. More importantly, we document consistent with Holmstrom’s informativeness principle how negotiations recalibrate initial target proposals to reflect peer performance on two performance signals. The relative weight assigned to one measure of peer performance versus the other will largely depend on its informational properties, that is how noisy each measure will be in estimating common shocks relative to the other. We conclude that by making extensive use of relative performance evaluation for target setting purposes managerial performance can be more precisely estimated by filtering out common shocks. As a result, well-performing managers can be more reliably rewarded with future economic rents by lowering their targets. This largely consistent with prior research which has suggested that commitment to existing contractual arrangements by not fully ratcheting up targets will help to preserve ex ante effort incentives.
Our study provides some managerial implications. From a contract efficiency perspective, our findings suggest that it might be beneficial for firms to equip local decision makers with decision rights to adjust headquarters’ initial target proposals. By doing so, targets adjustment can reflect an optimal set of performance signals and related incentive weights. Also, our study underscores that supervisors might benefit from a more elaborate performance measurement system that allows them to obtain a variety of information such as multiple performance measures or peer performance to evaluate managerial performance and set performance targets accordingly.
Our study is subject to a number of limitations. Most importantly, we provide singlefirm evidence from a rather specific field setting. Given the absence of earnings, the potential risk aversion of government bureaucrats and the organization specific target setting practice, it might be difficult to generalize our findings to other firms. However, research access to non-

37

perceptual company data is typically limited which may justify examining single firm settings as mine. A questionnaire-based survey across a larger number of firms might provide supplementary insights as to whether firms indeed rely on peer performance and possibly even on multiple signals thereof to set budget targets against which managers are evaluated and compensated at the end of the year. Furthermore, there is little empirical evidence from the field whether targets really matter with respect to performance. Therefore, future research might attempt to investigate whether and how target difficulty affects managerial performance. 38

References
Aggarwal, R. K., and A. A. Samwick. 1999. Executive Compensation, Strategic Competition, and
Relative Performance Evaluation: Theory and Evidence. The Journal of Finance 54 (6): 1999–
2043.
Albuquerque, A. 2009. Peer Firms in Relative Performance Evaluation. Journal of Accounting and
Economics 48 (1): 69–89.
Anderson, S. W., H. C. Dekker, and K. L. Sedatole. 2010. An Empirical Examination of Goals and
Performance-to-Goal Following the Introduction of an Incentive Bonus Plan with Participative Goal
Setting. Management Science 56 (1): 90–109.
Antle, R., and A. Smith. 1986. An Empirical Investigation of the Relative Performance Evaluation of
Corporate Executives. Journal of Accounting Research 24 (1): 1–39.
Banker, R. D., and S. M. Datar. 1989. Sensitivity, Precision, and Linear Aggregation of Signals for
Performance Evaluation. Journal of Accounting Research 27 (1): 21–39.
Bannister, J. W., and H. A. Newman. 2003. Analysis of Corporate Disclosures on Relative
Performance Evaluation. Accounting Horizons 17 (3): 235–46.
Black, D. E., S. S. Dikolli, and C. Hofmann. 2011. Peer Group Composition, Peer Performance
Aggregation, and Detecting Relative Performance Evaluation. Unpublished Paper, Duke University.
Bol, J. C., T. M. Keune, E. M. Matsumura, and Jae Yong Shin. 2010. Supervisor Discretion in Target
Setting:An Empirical Investigation. Accounting Review 85 (6): 1861–86.
Bouwens, J., and P. Kroos. 2011. Target Ratcheting and Effort Reduction. Journal of Accounting and
Economics 51: 171–85.
Bushman, R. M., R. J. Indjejikian, and A. Smith. 1996. CEO Compensation: The Role of Individual
Performance Evaluation. Journal of Accounting and Economics 21 (2): 161–93.
Campbell, D. 2008. Nonfinancial Performance Measures and Promotion-Based Incentives. Journal of
Accounting Research 46 (2): 297–332.
Courty, P., and G. Marschke. 2004. An Empirical Investigation of Gaming Responses to Explicit
Performance Incentives. Journal of Labor Economics, 23–56.
Covaleski, M., J. Evans, J. Luft, and M. D. Shields. 2007. Budgeting Research: Three Theoretical
Perspectives and Criteria for Selective Integration. In Handbook of management accounting research. Edited by C. Chapman, A. G. Hopwood, and M. D. Shields: Elsevier Science.
Cragg, M. 1997. Performance Incentives in the Public Sector. Evidence from the Job Training
Partnership Act. Journal of Law, Economics, and Organization 13 (1): 147–67.
Dikolli, S., C. Hofmann, and T. Pfeiffer. 2012. Relative Performance Evaluation and PeerPerformance Summarization Errors. Review of Accounting Studies (forthcoming).
Eckholm, B.-G., and J. Wallin. 2000. Is the Annual Budget Really Dead? European Accounting
Review 9 (4): 519–39.
Feltham, G. A., and J. Xie. 1994. Performance Measure Congruity and Diversity in Multi-Task
Principal/Agent Relations. Accounting Review 69 (3): 429–53.
Fisher, J., L. Maines, S. Peffer, and G. Sprinkle. 2002. Information Asymmetry on Budget Proposals,
Budget Slack, and Performance. The Accounting Review 77 (4): 847–65.
Gibbons, R., and K. J. Murphy. 1990. Relative Performance Evaluation for Chief Executive Officers.
Industrial and Labor Relations Review 43 (3): 30S.
Gong, G., L. Y. Li, and J. Y. Shin. 2011. Relative Performance Evaluation and Related Peer Groups in
Executive Compensation Contracts. Accounting Review 86 (3): 1007–43.

39
Hansen, S., D. Otley, and W. A. van der Stede. 2003. Practice Developments in Budgeting: An
Overview and Research Perspective. Journal of Management Accounting Research 15: 95–116.
Hansen, S., and W. A. van der Stede. 2004. Multiple Facets of Budgeting: an Exploratory Analysis.
Management Accounting Research 15: 415–39.
Holmstrom, B. 1979. Moral Hazard and Observability. The Bell Journal of Economics 10 (1): 74–91.
Holmstrom, B. 1982. Moral hazard in teams. The Bell Journal of Economics 13 (2): 324–40.
Holmstrom, B., and P. Milgrom. 1987. Aggregation and Linearity in the Provision of Intertemporal
Incentives. Econometrica 55 (2): 303–28.
Indjejikian, R. J., and M. Matĕjka. 2006. Organizational Slack in Decentralized Firms: The Role of
Business Unit Controllers. The Accounting Review 81 (4): 849–72.
Indjejikian, R. J., M. Matejka, K. Merchant, and W. A. van der Stede. 2011. Earnings Targets and
Annual Bonus Incentives. Unpublished paper, Arizona State University, Phoenix.
Indjejikian, R. J., and D. Nanda. 2002. Executive Target Bonuses and What They Imply about
Performance Standards. The Accounting Review 77 (4): 793–819.
Ittner, C. D., D. F. Larcker, and M. V. Rajan. 1997. The Choice of Performance Measures in Annual
Bonus Contracts. The Accounting Review 72 (2): 231–55.
Janakiraman, S. N., R. A. Lambert, and D. F. Larcker. 1992. An Empirical Investigation of the
Relative Performance Evaluation Hypothesis. Journal of Accounting Research 30 (1): 53–69.
Laffont, J., and J. Tirole. 1993. A theory of incentives in procurement and regulation. Cambridge,
Massachussetts: MIT Press.
Lambert, R. A. 2001. Contracting Theory and Accounting. Journal of Accounting and Economics 32
(1–3): 3–87.
Lambert, R. A., and D. F. Larcker. 1987. An Analysis of the use of Accounting and Market Measures of Performance in Executive Compensation Contracts. Journal of Accounting Research 25: 85–125.
Leone, A. J., and S. Rock. 2002. Empirical Tests of Budget Ratcheting and its Effect on Managers’
Discretionary Accrual Choices. Journal of Accounting and Economics 33 (1): 43–67.
Libby, T., and R. Lindsay. 2010. Beyond Budgeting or Budgeting Reconsidered? A Survey of NorthAmerican Budgeting Practice. Management Accounting Research 21: 56–75.
Mankiw, G. 2008. Macroeconomics. 6.th ed. New York: Worth Publishers.
Matsumura, E. M., and J. Y. Shin. 2006. An Empirical Analysis of an Incentive Plan with Relative
Performance Measures: Evidence from a Postal Service. The Accounting Review 81 (3): 533–66.
Milgrom, P., and J. Roberts. 1992. Economics, organization and management: Englewood Cliffs,
United States : Prentice-Hall.
Murphy, K. J. 2001. Performance Standards in Incentive Contracts. Journal of Accounting and
Economics (30): 245–78.
Ouchi, W. G. 1979. A Conceptual Framework for the Design of Organizational Control Mechanisms.
Management Science 25 (9): 833–48.
Ouchi, W. G. 1980. Markets, Bureaucracies, and Clans. Administrative Science Quarterly 25 (1): 129–
41.
Phillips, P. C. B., and J. Y. Park. 1988. On the Formulation of Wald Tests of Nonlinear Restrictions.
Econometrica 56 (5): 1065–83.
Rogers, W. 1993. Regression Standard Errors in Clustered Samples. Stata Technical Bulletin 3 (13):
19–23.
Shields, J., and M. Shields. 1998. Antecendents of Participative Budgeting. Accounting, Organizations and Society 23 (1): 49–76.

40
Umapathy, S. 1987. Current budgeting practices in U.S. industry: The state of the art. New York, NY:
Quorum Books.
Weitzman, M. L. 1980. The "Ratchet Principle" and Performance Incentives. The Bell Journal of
Economics 11 (1): 302–08.
Zellner, A. 1962. An Efficient Method of Estimating Seemingly Unrelated Regressions and Tests for
Aggregation Bias. Journal of the American Statistical Association 57 (298): 348–68.