A Catering Theory Of Dividends

A Catering Theory of Dividends

Baker, Malcolm Wurgler, Jeffrey.

The Journal of Finance VOL. LIX (2004) 1125-1165


What we know about dividend policy so far? Obviously, there are a lot of theories or empirical evidence, but results are mixed. The majority extant literature has an implicit assumption of efficient market, and this is exactly what Baker and Wurgler want to debate on. In doing so, Baker and Wurgler provide a catering theory. The basic idea, is that since limits on arbitrage fail to eliminate some investors’ demand for cash-dividend-paying firms and such a demand affects current share prices, managers rationally take advantage of the benefit from current mispricing net of long-run cost and make dividend payment decision accordingly. The empirical implication of Catering Theory is that the propensity to pay dividends depends on a dividend premium (or sometimes discount) in stock prices. We need to be cautious here, this paper is about the discrete decision whether to pay dividends, not how much to pay. As a paper in behavioral finance, this paper is related to Shefrin Statman [1984], in which self-control, prospect theory and regret aversion are listed as three psychological explanations about why some investors prefer dividends.



The model presented above predicts that the propensity to pay dividends is increasing in dividend premium, which displays its variation in the time series. Therefore, feasible empirical test is to examine the ability of proxies for the dividend premium to explain time series variation in the rates of dividend initiation, continuation, and omission.

The research time frame is from 1962 to 2000. Data are from COMPUSTAT & CRSP (shared codes of 10 or 11). Baker and Wurgler excluded stocks with book equity below $250,000 or assets below $500,000 and utilities and financial firms.

Baker and Wurgler defined Dividend Payment Variables as follows:
Summary statistics is on Table I right column. It shows that Initiation rate drops in late 1960, rebounds mid 1970s, drops again late 1970s, remains low ever since. Continuation rate is stable, and List-Pay rate declines overtime.

Baker and Wurgler also defined four Dividend Premium Variables as follows:
Dividend Premium It is measured as log of the ratio of average (equal-weighted and value-weighted specification) market-to-books for payers and non-payers. Summary statistics is provided in Table II & figure 1. It shows that Dividend premium varies over time and it coincides with the change of dividend payment measure (recall Table I). The disadvantage of using this measure is that it doesn’t control for other characteristics, such as size and profitability, and it could also reflect relative investment opportunities.
CU dividend premium It is measured as log of the ratio of cash payout share’s price and stock payout share’s price. Summary statistics is provided in Table III. It shows that CU dividend premium also varies over time. The advantage of using this measure is that it doesn’t reflect aggregate investment opportunities; however, the disadvantage is that it provides a long-horizon arbitrage opportunity, it is only for one firm, and this experiment ends in 1990.
Average announcement effect of recent initiation The intuition of using this measure is that if investors favor dividend, they would react positively. Announcement effect of recent initiation is measured as average 3-day event window CAR with respect to CRSP valuated weighted index. After obtaining CAR for individual firm, Baker and Wurgler created a standardized cumulative abnormal announcement return A. Summary statistics is provided in Table III. We can observe that all measures of A, with only one exception, are positive, some of them are significant.
Future Returns The intuition of using this measure is that if managers indeed rationally initiate dividends to exploit market mispricing, a high rate of initiations should forecast low returns on payers relative to non-payers as the relative overpricing of payers reverse. Future returns are defined as difference between of payers and non-payers, in terms of future returns (1-year and 3-year accumulation specification) on value-weighted indexes.
The correlation of these dividend premium measures is provided in Table IV. It is shown that Dividend premium, CU premium and announcement effect are positively correlated; future returns are negatively correlated with other measures. These results are consistent with intuition.

The visual evidence of time series relationship is in Fig 2. It shows that dividend premium and raw rate of dividend initiation in the following year is positively related, esp. it the first part of the time period. The relation breaks down in the second part. Baker and Wurgler argued that this can be attributed to the influx of small, high growth, less profitable firms on the exchange, as recorded in Fama French [2001].
The first regression specification is following:
The results are provided in Table V. It is shown that dividend premium, CU premium and announcement reaction all carry statistically significant sign as predicted by Catering Theory in univariate regression, but dividend premium best captures the common factor.
The second regression specification is following:
The results are provided in Table VI. It is shown that dividend decision can predict future relative return, but not absolute return. We can observe this by looking at p-value. In the regression with absolute return as dependant variable, the variables of interest are not significantly different from zero.


Baker and Wurgler first ruled out the possibility that the results are driven by “Common Time Trend” . They include year dummy in the regression, but the coefficient on dividend premium is still significant after including year dummy (Table VII).

Since the primary variable, dividend premium can also reflect investment opportunities, and the results could be driven by time varying investment opportunities. In order to rule out this possibility, Baker and Wurgler used different approaches. First, they control for alternative proxy of investment opportunities (average market-to-book ratio or CRST value-weighted dividend yield) in multivariate regression. The results provided in Table VII shows that dividend premium is still significant after control, thus it suggests that investment opportunities don’t appear to drive the results. Oil industry had big investment opportunities in early 1970s. If dividend premium is a proxy for investment opportunities and the results are driven by investment opportunities consequently, we should expect to observe a negative decline in the number of dividend payers in oil industry, but Baker and Wurgler found the opposite results. Moreover, unreported results show that the correlation between rate of repurchase and lagged dividend premium is insignificant and negative. Lastly, unreported results show that neither the payout ratio nor the dividend yield is significantly correlated with the lagged dividend premium. This confirms that lagged dividend premium is only correlated with the rate of dividend initiation, continuation, and omission.

Baker and Wurgler also ruled out the alternative explanation about “Correlated Errors in Forecasting Investment Opportunities”. If it is true, it can only reinforce the result.

Clearly, the primary variable, dividend premium also reflects other firm characteristics, and the time series analysis results could possibly be driven by the “Time-Varying Characteristics”. In order to rule out this explanation, Baker and Wurgler run a 2SLS regression. In the first stage, a residual propensity to initiate dividend payment is obtained from a logit regression after controlling for other characteristics. In the second stage, the aggregated residual propensity to initiate dividend payment is regressed on dividend premium. The basic idea is that if dividend premium continues to show significance, Catering Theory holds. The regression specification is follows:
The results are provided in Table VIII & fig 3. It shows that controlling for firm characteristics doesn’t change the result.

Lastly, Baker and Wurgler ruled out another possible explanation, “Time-Varying Contracting Problems”. If dividend payout is a means to curb agency cost, it should be negatively related with corporate governance. However, this explanation requires an explanation of governance improvement in 1960s when the propensity of dividend payout was declining.

Managers are not necessarily smart in chasing mis-pricing. They may just cater to, or even be force by proxy vote to meet, extreme investor demands in general, and mispricing is merely a symptom of extreme investor demand. This is a softer view of catering.

Baker and Wurgler muted other explanations about “Catering to Clienteles”. First, if the dividend the dividend premium were being driven by changes in the structure of rational clienteles, it should have a closer connection to the level of dividends than to the number of payers. Baker and Wurgler found the opposite. Secondly, Baker and Wurgler controlled tax effect in their multivariate regression. The results reported in Table VII shows that the added tax control variable doesn’t impact dividend premium too much. Unreported results also rule out the possibility that transaction cost drives the results. Lastly, Baker and Wurgler also ruled out institutional investor clientele explanation, since institutional ownership has rose since 1980s, and it is hard to reconcile the time-varying pattern in 1960s.

Finally, Baker and Wurgler provided their story: Catering to Investor Sentiment.
Visual Evidence is provided in Fig 4: there is positive relationship between dividend premium and close-end funds discount which has been documented in prior literature as sentiment measure (Lee et al [1991]).
Baker and Wurgler use close-end funds discount as instrument to further conduct a 2SLS regression. The regression specification is following:
Both 2SLS and OLS results are reported in Table IX. 2SLS results are as strong as those of OLS, supporting the sentiment interpretation, to the extent that all the IVs pick up the investor sentiment.


  • Catering Theory: managers tend to initiate dividends when investors put a relatively high stock price on dividend payer, and tend to omit dividends when investors prefer non-payers.
  • Empirical Results are broadly consistent with this catering Theory.
  • Sentiment appear to be a key factor to explain the source of investor demand for dividends.
  • Caution: Catering explains the number of payers but not the total payouts by existing payers. It is about whether to pay dividends, not how much to pay. Once dividends are initiated, increases and decreases appear to be governed more by firm-level profitability than by relative valuations.
Unless otherwise stated, the content of this page is licensed under Creative Commons Attribution-ShareAlike 3.0 License