When is good news really good news?

Abstract The impression management literature suggests that managers often resort to biased disclosures. However, there is little systematic evidence on what types of strategies management uses to achieve this bias. Do managers simply lie? Or, do they use more subtle ways of introducing positive bias into corporate narratives, such as selecting specific information items which result in a more positive impression ('selectivity') or by keeping their narratives vague and general ('vagueness')? In order to differentiate between the two scenarios, I re-examine the positive forward-looking statements examined by Schleicher and Walker (2010) and compare, across firms with improving and deteriorating financial performance, the managerial choices made in relation to eight forecast attributes. I make two observations. First, there are significant differences in the characteristics of good- and bad-news firms' positive statements. In particular, bad-news firms' positive statements involve more non-specific time horizons, more segmental forecasts, and more references to conditions and aims and objectives, but fewer directional forecasts, fewer numbers, and fewer reinforcing qualifiers. Second, the identified differences in good- and bad-news firms' positive statements can be exploited for classification purposes: including into a classification model additional regressors that measure a positive forward-looking statement's level of selectivity and vagueness significantly increases the model's ability to separate firms with improving financial performance from firms with deteriorating financial performance. Overall, my results are consistent with (a) impression management operating predominantly through selectivity and vagueness and (b) selectivity and vagueness being an important signal for future financial performance. Keywords: impression managementnarrativesselectivitysignallingvagueness Acknowledgement The author is grateful to the editor and two anonymous reviewers for helpful comments on earlier versions. He would like to thank Martin Walker for providing a second set of manual disclosure scores. The financial support from the ESRC, Grant Number RES-000-22-1089, is gratefully acknowledged. Notes I am grateful to an anonymous referee for pointing out that a significant increase in classificatory power might also be consistent with the predictions from a 'modified' 'cheap talk' model. In particular, in 'cheap talk' models (e.g. Crawford and Sobel 1982 Crawford, V. and Sobel, J. 1982. Strategic information transmission. Econometrica, 50(6): 1431–1451. [Crossref], [Web of Science ®] , [Google Scholar]), managers are assumed to be rational in pursuing their own personal gains, but their signals are biased, as their preference system is not aligned with those of investors. Furthermore, as investors are rational and understand the manager's motivation, they discount the biased (or uninformative) (positive) signal. The 'modification' then comes in when investors use selectivity and vagueness instead for classification purposes. I discuss other possible explanations for a significant finding in the conclusion. Note that my keyword lists are intended for general guidance only. Thus, while it is hoped that the presence of keyword lists helps to increase the consistency of the coding process over time, they are not 'binding' in the sense that the inclusion (or non-inclusion) of a certain keyword automatically leads to a specific classification. This is particularly true as the keyword lists are unlikely to be an exhaustive list of all relevant keywords. Thus, ultimately, it is the coder's interpretation of the underlying meaning – not the presence of a keyword – that determines the classification. Finally, note that not all forecast attributes are supported by keyword lists. For example, while positive and negative impressions are supported by a list, directional forecasts (and performance indicators) are not. To illustrate the underlying calculations, I assume that a firm makes the three positive statements in Table 1, one for 'sales', one for 'earnings', and one without an explicit performance indicator. Such a firm would receive a score of 0.33 each under 'sales', 'earnings', and 'general unspecified statement'. This is calculated by dividing the number of statements per performance indicator over the total number of statements. Also, such a firm would score 0.66 for 'directional', 0.33 for 'non-directional', 0.33 each for 'quarter', 'year', and 'no time horizon', and 0.33, 0.66, and 0.33 for 'quantitative' 'reinforcements', and 'conditional statement', respectively. All other variables in Table 1 would be zero. Note that the p-values in Table 2 are two-tail test p-values even though the alternative hypotheses involve one-sided predictions. Thus, the p-values in Table 2 are conservative and should be interpreted as a lower bound in terms of significance levels. In particular, a 10% p-value in Table 2 translates into a 5% one-tail test p-value. Note that the values in Table 3 – unlike the values in Table 2 – are derived from variables that follow a binomial distribution. Also, note that Table 3 – like Table 2 – only reports p-values from a parametric two-sample t-test. However, untabulated p-values from a non-parametric Wilcoxon rank-sum test and a binomial proportion test are very similar. There are two reasons for why I do not further analyse negative (and neutral) statements. First, the prior literature, including Clarkson et al. (1994 Clarkson, P. M., Kao, J. L. and Richardson, G. D. 1994. The voluntary inclusion of forecasts in the MD&A section of annual reports. Contemporary accounting research, 11(1): 423–450. [Crossref] , [Google Scholar]) and Schleicher and Walker (2010), Schleicher, T. and Walker, M. 2010. Bias in the tone of forward-looking narratives. Accounting and business research, 40(4): 371–390. [Taylor & Francis Online], [Web of Science ®] , [Google Scholar] finds that negative tone statements, unlike positive tone statements, are highly ex post accurate, presumably because there are much fewer managerial incentives to disclose negative statements when the outlook is indeed positive. Thus, it seems that negative statements, unlike positive statements, can be taken at face value without a need for further analysis. Second, the number of negative (neutral) statements in UP is probably too small to empirically detect significant differences between UP and DOWN, even if such differences existed. –2LL is often called the 'likelihood ratio'. It is a goodness-of-fit statistic for logistic regressions and measures the unexplained variance in the dependent variable, analogous to the sum of squared errors in OLS (Garson 2011 Garson, D. J. 2011. Logistic regression [online]. North Carolina State University. Available from: http://faculty.chass.ncsu.edu/garson/PA765/logistic.htm [Accessed 25 August 2011] [Google Scholar]). The sign on POS (NEG) is expected to be positive (negative) as the presence of a positive (negative) outlook statement is expected to increase the odds of a positive (negative) financial performance over the coming year. The coefficient on NEU is more difficult to predict but – given the evidence in Schleicher and Walker (2010 Schleicher, T. and Walker, M. 2010. Bias in the tone of forward-looking narratives. Accounting and business research, 40(4): 371–390. [Taylor & Francis Online], [Web of Science ®] , [Google Scholar]) – I expect the coefficient to be negative. The four impression management dummies – DIRECTIONAL j , QUARH1YEAR j , QUANREIN j , and NOTSEGCONAIM j – are set equal to 0 if selectivity and vagueness are present and this presence is expected to increase the likelihood of a negative financial performance in the coming year. Thus, I predict a positive sign on DIRECTIONAL j , QUARH1YEAR j , QUANREIN j , and NOTSEGCONAIM j . In contrast, the expected sign of the coefficient on PI j varies with the nature of the underlying performance indicator. In particular, I expect a positive sign for 'sales' and 'earnings' and a negative sign for all other performance indicators. Finally, note that I do not predict any coefficient signs for financial statement variables as the association between financial statement variables and future financial performance is often ambiguous. For example, are higher inventory turnovers a signal of increased efficiency, or are they a result of reduced inventory levels in anticipation of an expected slow-down? Note that many of the 69 variables in Appendix 2 measure similar economic constructs and are likely to be highly correlated. Thus, it is not surprising that changes in the outlier definition lead to changes in the list of significant financial statement control variables. Also, I do not winsorise the three tone variables as they are not affected by extreme outliers: in the full model POS, NEU, and NEG have minimum (maximum) values of 0, 0, and 0 (7, 4, and 4) and means (medians) of 2.17, 0.47, and 0.79 (2, 0, and 1). Finally, the impression management variables are 0/1 dummies. In the full model, the average mean for PI j , DIRECTIONAL j , QUARH1YEAR j , QUANREIN j , and NOTSEGCONAIM j is 0.25, 0.12, 0.10, 0.06, and 0.17. The coefficient on QUARH1YEAR j in the last column of Table 5 is negative, inconsistent with my prior results and my prior expectation. It is important to remember, however, that the coefficient of –3.16 represents the average coefficient from the underlying six dummies, namely the dummy for 'sales', 'costs', 'earnings', 'trading', 'general unspecified statements', and 'industry'. These six dummies have coefficients of –3.78, –19.45, 1.60, 2.31, –0.46, and 0.82. Clearly, the value of –3.16 is influenced by the large negative coefficient for QUARH1YEARCOSTS which, in this regression, is estimated from only seven non-zero observations. In contrast, the median value for QUARH1YEAR j of 0.18 is positive. Finally, note that in all but three cases, the average coefficient in Tables 4 and 5 always has the same sign as the untabulated median coefficient. The two other exceptions apply to PI j and DIRECTIONAL j in the regression with outliers defined at the 1% and 99% level. It is well known that the European Central Bank, ECB, signals interest rate changes to the market well ahead of formal interest rate decisions. However, rather than signalling the impending change in quantitative terms, it uses qualitative keywords – i.e. 'coded' messages – to reveal its intentions. The third explanation implies that UK-listed companies behave in a way similar to the ECB. Clearly, the third explanation implies that selectivity and vagueness do not entail opportunistic managerial behaviour with the intention to mislead. In particular, the tendency of companies to present themselves in the best possible light would not be a result of impression management, but of corporate reporting conventions. It is in this sense that the third explanation, if correct, would require a rethink in the impression management literature. At the same time, there would no longer be a need 'to reconcile the information content school with the impression management literature'.