The Stewardship Role of Analyst Forecasts, and Discretionary Versus Non-discretionary Accruals

Abstract

We examine the interaction between discretionary and non-discretionary accruals in a stewardship setting. Contracting includes multiple rounds of renegotiation based on contractible accounting information and non-contractible but more timely non-accounting information. We show that accounting regulation aimed at increasing earnings quality from a valuation perspective (earnings persistence) may have a significant impact on how firms rationally respond in terms of allowing accrual discretion in order to alleviate the impact on the stewardship role of earnings. Increasing the precision of more timely non-accounting information (analyst earnings forecasts) increases the ex ante value of the firm and reduces costly earnings management. There is an optimal level of reversible non-discretionary accrual noise introduced through revenue recognition policies. Tight rules-based accounting regulation, as opposed to leaving firms more choice over non-discretionary accrual policies, may lead firms to rationally respond by inducing costly earnings management. More generally, regulating both earnings persistence and the tightness of admissible auditing policies may not result in less equilibrium earnings management.

Notes

In his presidential address to the 2003 AAA annual meeting, Demski 2004 also stresses that the impact of changes in accounting regulation cannot be addressed without considering how firms and market participants may react to these changes.

Note that these comments are merely statements about increased information content of earnings, not about whether this is valuable to shareholders (and others). In reasonably perfect capital markets, one must work very hard to demonstrate the latter. Diversification is a strong force here (see Feltham and Christensen, 1988).

Representing high earnings quality by high earnings persistence assumes that earnings persistence is measured over the lifetime of the firm. For example, smoothing earnings such that the same earnings number is reported every year yields perfectly correlated earnings, but also earnings carrying no intertemporal information. Articulation, however, implies that realized aggregated cash flows must be revealed through last-period earnings.

Christensen and Feltham (2005, Chapter 28) show in the Corollary to Proposition 28.7 that, with fully optimal contracts, renegotiation forces the contract to be additively separable in first- and second-period earnings innovations which, in turn, eliminates the ex ante covariance incentive.

Within a single-period agency model Hermalin and Katz (1991) show that if the early non-contractible information perfectly reveals the agent's action, then first-best can be achieved: effectively, the agent's action becomes contractible through the renegotiation encounter. At the other extreme, Fudenberg and Tirole (1990) show that if the non-contractible information is completely uninformative about both the agent's past action and the forthcoming contractible performance measure, i.e., the signal merely reveals that the agent has taken his action, then the only equilibrium in pure strategies is one in which the agent gets a fixed wage and provides the least cost level of effort. Christensen et al. (2002) extend the latter model to a multi-period setting with uncorrelated earnings, and show that allowing the manager to costlessly build ‘secret reserves’ makes the shareholders better off. Allowing ‘secret reserves’ implicitly commits the shareholders to maintain a contract with a positive slope on first-period earnings after renegotiation (since, otherwise, the manager would shift all income into the second period) – although, in equilibrium, no ‘secret reserves’ are built.

Dutta and Gigler (2002) also consider the impact of earnings forecasts. In their model, the earnings forecast is a voluntary disclosure of private information by the manager, while we assume the earnings forecast comes from analysts or, more generally, represents one of several public and private sources of non-contractible performance information. Gigler and Hemmer (2004) also have a public earnings forecast that is contractible through renegotiation, but are concerned with comparing its value to having the manager report a privately observed earnings forecast (and allowing earnings management). In contrast, we are interested in the impact of the public forecast on earnings management, so we have both in the model rather than as alternative reporting regimes.

Compare wit Liang (2004) and ‘window dressing’ in Feltham and Xie (1994) and Dutta and Gigler (2002).

We consider a setting in which management has no new information about future cash flows so that more aggressive revenue recognition does not have a countervailing effect of increasing earnings persistence through recognition of news about future cash flows. We leave this latter effect of revenue recognition for future research.

With ‘fair value’ accounting, time-series variation in ROE is transitory noise, whereas earnings has transitory, reversible, and persistent components. Note that with ‘fair value’ accounting, the accounting ROE equals the shareholders' market rate of return.

Golden parachutes and deferred compensation provide commitment to multi-period employment. We also assume the principal's expected benefit from operating the company is large enough that he will never terminate operations.

We assume the principal can commit to not renegotiating the auditing tightness parameter α, which is consistent with the choice of auditor (strict or lax) for the following year being made after the annual financial report at the general assembly. If we were to allow for renegotiation of α after the first-period action has been taken and the leading indicator is observed, the renegotiated contract would be based on α = 0 such that no costly earnings management would occur in equilibrium. Of course, anticipating such renegotiation would imply that earnings management cannot be used to alleviate the first-period incentive problem with respect to .

The results are qualitatively no different if we assume time-additive agent utility with unrestricted borrowing and saving, and an infinite consumption horizon, see Dutta and Reichelstein (1999), Christensen et al. (2003a), and Şabac (2007). Since consumption smoothing by the agent is not an issue, we use the simpler model.

Note that earnings management is a mean effect only and, therefore, it does not affect conditional variances. Consequently, we express the conditional variance in the incentive ratio in terms of the true performance measures and .

If , as determined by equation (9), then the incentive rate under is identical to the incentive rate that will be offered at R2. Hence, as the only remaining uncertainty pertains to , the principal gains only if the agent accepts a lower continuation expected utility under at . Obviously, the agent rejects such offers and, hence, any contract accepted at , for which , is renegotiation-proof at R2.

If there is no earnings management, and if the performance measures are uncorrelated given , then, similar to Fudenberg and Tirole (1990), .

Of course, if earnings management is costless, must be equal to in order to avoid infinite expected utility to the agent.

In CFS it is never valuable to allow costly earnings management by the agent, provided the principal can preclude earnings management through a costless audit technology.

In a setting similar to the zero correlation case, Christensen et al. (2002) show that the choice of discretionary accounting policies can mitigate the problem of inducing zero first-period effort by allowing the agent to build secret reserves. This forces the principal to choose a positive incentive rate on the first-period performance report.

Note that the optimal interior α does not depend on the principal's marginal benefit of agent effort, . This holds more generally whenever .

The leading indicator is a noisy forecast of and has no value with full commitment, even if it were contractible. Here, the leading indicator is valuable due to its role in motivating productive first-period effort.

The under-reporting result is a consequence of our focus on the identical periods case. In a more general model, under-reporting as well as over-reporting may occur.

See, for example, Dechow and Dichev (2002). In their framework, accrual estimation errors can come from either earnings management or from accounting policies. In our case, earnings management is noiseless income shifting by the agent and only accounting policies are associated with accrual estimation errors. In this respect, our discretionary accruals are noiseless, and only the non-discretionary accruals are noisy.

In our model there is no difference between pre- and post-earnings management persistence because the agent's earnings management does not depend on his information and is a non-random action that can be rationally inferred in equilibrium.

Consider a modification of the model in Fudenberg and Tirole (1990), where a forecast of the contractible performance measure is observed prior to renegotiation. The principal uses the forecast to provide implicit incentives for productive effort and, in the limit, when the forecast perfectly reveals the performance measure, the solution without renegotiation is achieved. Alternatively, assume the early signal partially reveals the performance measure's noise. This limits the principal's ability to use the performance measure to provide insurance for other compensation. In the limit, when the forecast perfectly reveals the noise in the performance measure, the principal achieves the solution without renegotiation.

Additional information

Notes on contributors

Peter O. Christensen

Paper accepted by Laurence van Lent.

Hans Frimor

Paper accepted by Laurence van Lent.

Florin Şabac

Paper accepted by Laurence van Lent.