Window Dressing by Pension Fund Managers

by Josef Lakonishok, Andrei Shleifer, Richard Thaler, Robert Vishny
Citation
Title:
Window Dressing by Pension Fund Managers
Author:
Josef Lakonishok, Andrei Shleifer, Richard Thaler, Robert Vishny
Year: 
1991
Publication: 
The American Economic Review
Volume: 
81
Issue: 
2
Start Page: 
227
End Page: 
231
Publisher: 
Language: 
English
URL: 
Select license: 
Select License
DOI: 
PMID: 
ISSN: 
Abstract:

The conventional wisdom on Wall Street is that portfolio managers are reluctant to produce annual reports that show holdings of shares that have sharply declined in val- ue. One money manager is quoted by S. Jansson as saying: "Nobody wants to be caught showing last quarter's disasters.. . . You throw out the duds because you don't want to have to apologize for and defend a stock's presence to clients even though your investment judgment may be to hold" (1983, p. 139). The practice of deleting such shares from portfolios at year end is an example of "window dressing." Despite conventional wisdom, little is known of how widespread this practice is. In this paper, we look for evidence of window dressing among an im- portant group of portfolio managers, those who manage pension funds.

As of December 1989, pension funds owned close to $900 billion of American equities, about 25 percent of the stock mar- ket capitalization. Of this amount, about $700 billion is actively managed and the rest is invested in index funds. Some plans man- age their money internally, but more com- monly they hire several money managers and split the pension plan's money among them. Plan sponsors typically evaluate fund managers once a quarter, but the main eval- uation takes place at the end of the year. Based on these evaluations, assets are real- located across money managers. Window dressing, if it occurs, is presumably a response to these evaluations.

Why might a fund manager engage in window dressing? The main focus of fund

*University of Illinois, Champaign, IL 61820; Har- vard University, Cambridge, MA 02138; Cornell Uni- versity, Ithaca NY 14850; and University of Chicago, Chicago, IL 60637; respectively. We are grateful to Gil Beebower and Vasant Kamath for help with this project.

evaluations is past performance relative to some benchmark, such as the S&P 500. Stock returns, however, are noisy and may not suffice to identify the manager's invest- ment philosophy or to see whether he just got unlucky on a few stocks. As a result, sponsors may look at the actual portfolio holdings as well. To impress sponsors, fund managers may alter these portfolios at the end of the quarter, and especially at the end of the year, that is, window dress. A finding that money managers window dress is there- fore evidence that managers are evaluated on a broader set of criteria than perfor- mance alone.

Selling losers is the most frequently men- tioned form of window dressing. Pension funds can also reduce the pace of sale of winners, increase the purchases of winners and reduce the purchases of losers to im- press the sponsors with the looks of their portfolios.' Some of these strategies, how- ever, may not be as effective as selling losers. When a fund buys winners after they rose in price, sponsors would realize that these win- ners were not held during the price rise. If a fund purchases losers, its manager can probably explain that he bought them after they had fallen. On the other hand, slowing down the sale of winners might be as attrac- tive as speeding up the sale of losers. We will examine these less frequently mentioned and perhaps less sophisticated forms of window dressing as well as the generic dumping losers strategy.

'R.Haugen and Lakonishok (1988) suggest window dressing as a possible explanation of the "January Effect." Lakonishok and S. Smidt (1988) find a very substantial increase in the Dow Jones Industrial Aver- age (an index of 30 large stocks) in the last week of December. They offer window dressing as a possible explanation of this finding.

I. The Data

Our analysis is based on a proprietary sample of 769 equity pension funds pro- vided by SEI, a professional fund evaluation service. Equity funds hold at least 90 per- cent of their money in equities. The data set contains, for each fund, complete equity portfolio holdings at the end of each quar- ter from 1985 to 1989. Sponsors of most of these funds are corporations, but there are some state, municipal, and endowment pen- sion funds as well. The total amount under management in these 769 funds at the end of 1989 is $124 billion, or about 18 percent of the total actively managed holdings of pension funds. The average equity holdings of a fund at the end of 1989 are $161 million; the largest 5 percent of the funds manage 65 percent of the money.

We do not observe all the trades that a fund makes in a given quarter because a fund might have bought and sold the same stock within that quarter. However, we can estimate purchases and sales based on port- folio changes from quarter end to quarter end. In the sample, in an average quarter, a fund buys 13 percent of the dollar value of its previous quarter's holdings, and sells 12 percent. Purchases exceed sales in part be- cause funds receive dividends and in part because new contributions to the funds from the sponsors typically exceed payouts to the retirees. The turnover rate of 50 percent a year is typical for institutions. Pension funds' holdings are heavily concentrated in the largest capitalization stocks. While the bot- tom 50 percent of stocks in terms of size comprise 4.3 percent of market capitaliza- tion, they only represent 1.4 percent of pen- sion funds' equity holdings.

To determine whether managers engage

in window dressing, we must classify stocks

into performance categories. At the end of

each quarter, we take the CRSP universe of

all NYSE, American, and NASDAQ stocks,

and divide it into equal size quintiles based

on stock returns over the past year up to the

end of that quarter. This procedure is re

peated every quarter, so the composition of

each quintile changes each quarter. Since

we know which stocks belong to which quin-

Past Performance
Quintile

1 2 3 4 5
Universe Holdings in        
Quintile as % of Total        
Universe Holdings .06 .16 .23 .28 .28
Fund Holdings in          
Quintile as % of          
Total Fund Holdings .05 .17 .25 .29 .24
Sales in Quintile          
as % of          
Total Sales .06 .17 .22 .27 .28
Purchases in Quintile          
as % of          
Total Purchases .06 .19 .24 .28 .23

'All numbers are averaged over quarters and years.

tile, we can compute how much each fund holds, buys, and sells each quarter from each quintile. We use the average of begin- ning and end-of-quarter stock prices to esti- mate the total value of purchases and sales in each quintile by each fund.

Table 1 presents the fractions of their stock portfolios that funds as a whole hold in each past performance quintile (averag- ing over years and quarters) and the frac- tion of value of the CRSP universe in those quintiles. Table 1 also reports the fraction of purchases and sales that the funds as a whole make in each quintile. Holdings, sales, and purchases are all computed as if all money was in one aggregate fund. Table 1 shows that funds hold less of extreme losers and winners than exist in the universe. Funds hold only 5 percent of their portfo- lios in extreme losers, compared to 6 per- cent for the universe portfolio. They also hold 24 percent of their portfolios in extreme winners, compared to 28 percent for the universe. Obviously, funds are over represented in holdings of intermediate per- formance quintiles. One reason for the underrepresentation in extreme performance categories is that funds hold few small stocks, which are more likely to be among the extreme losers and extreme win- ners.

11. Activity of Pension Funds by Quarter

All accounts of window dressing stress the relative importance of the year-end re- port. This section compares trading in the first three quarters with that in the fourth.

About 20 percent of the funds in an aver- age quarter have zero holdings in the bot- tom performance quintile. To avoid statisti- cal problems this raises when we examine the cross section of funds, we combine the bottom two quintiles, so the bottom group in a quarter is 40 percent worst performing stocks over the previous year. As a result of combining quintiles, funds holdings are more evenly distributed across the four groups.

Our measure of selling intensity in per- formance group i in quarter j by fund k is

where SELL(i, j, k) is the value of sales by fund k in quarter j and group i, and HOLD(i, j -1, k) is the value of holdings at the end of the previous quarter of the exact same stocks as those in performance group i in quarter j. Both SELL and HOLD are defined using the average of the beginning and end of quarter j prices. The numerator of (1) is the ratio of the sales in a perfor- mance group to holdings of the same stocks at the end of the previous quarter. The denominator is the ratio of total sales in this quarter to holdings at the end of the previ- ous quarter.

Equation (1) measures the selling inten- sity of a fund in a given performance group relative to the overall selling intensity. It corrects for the fact that funds might be selling less in a given group at a given time only because they are selling less of every- thing. If a fund sold 50 percent of the extreme winners it holds, but only 20 per- cent of its total holdings, its sales index for winners according to (1) is 2.5. Having com- puted this number for every fund, quarter and year, we average it over funds, years, and quarters 1-3, and then compare the average over quarters 1-3 to that for quar- ter 4.

Our measure of buying intensity by fund k in quarter j in performance group i is

where BUY(i, j, k) is dollar purchases by fund k in group i in quarter j and UNIV.HOLD(i, j) is the value of CRSP universe holdings in quarter j in group i. Again, both variables are computed using the average of beginning and end of quarter j prices. This number is the fraction of purchases that a fund makes in a performance category relative to the fraction of universe holdings in that performance cate- gory. So if 20 percent of a fund's purchases were in performance group 4, which is only 10 percent of the universe value in that quarter, the number for the fund in that quarter and quintile is 2.

Equation (2) controls for the value of each performance group that a fund can buy in every quarter, so purchases are scaled by the availability of stocks to buy. In this respect, the numbers for sales and purchases are parallel: the universe holdings define availability of stocks for purchase, and since funds in general do not take short positions, their own holdings define the availability of stocks for sale. We compute

(2) for each fund, and then, as with measure

TBLE 2-SALES AND PURCHASES

OF

PENSIONFUNDSBY QUARTER^

Quintile Quarter 1-3 Quarter 4 t-~est~

A. Sales Relative to Own Holdings

1-2 1.32 1.42 1.75
(.026) (.033)
3 .92 1.02 1.76
(.009) (.084)
4 .91 .89 .24
(.007) (.014)
5 1.44 1.37 1.22
(.053) (.028)

B. Purchases Relative to Universe Holdings

1-2 1.22 1.33 5.47 (.012) (.024)

  1. 3 1.07 1.09 .097 (.009) (.017)
  2. 4 1.01 1.00 .44 (.008) (.013)
  3. 5 .90 .88 .71 (.009) (.015)

asales number is from equation (I), purchase num- ber is from equation (2). Both are averaged over funds and years.

or difference, Quarter 1-3 = Quarter 4.

(11, average over funds and years for each quarter, and finally average over quarters 1 through 3 and compare that average to the one for quarter 4.

Table 2 presents the results, with cross-sectional standard errors for each quarter- performance group mean (in parentheses) and t-tests of equality of means for quarters 1-3 on the one hand and quarter 4 on the other. Unlike the results in Table 1 that were computed as if all holdings and trades were by one aggregate fund, in Table 2 we compute measures (1) and (2) first for each fund and then average over funds. This en- ables us to compute cross-sectional standard errors.

To begin, consider the findings for quar- ters 1-3. First, when it comes to purchases, funds are clearly contrarian: relative to availability they overbuy losers (ratio of 1.22) and underbuy winners (ratio of .90). Second, when it comes to sales funds oversell winners relative to their holdings (ratio of 1.44), but they also oversell losers (ratio of 1.32). Much less selling activity is observed in the middle quintile groups. These results confirm Table 1's findings that funds are generally contrarian, except that they over- sell extreme losers, consistent with the win- dow-dressing hypothesis.

The comparison across quarters shows that the sale of losers relative to other quar- ters does accelerate in the fourth quarter, as it should if fund evaluations are most critical at the end of quarter four. In quar- ters 1-3, an average fund sells 32 percent more of the extreme losers (relative to its holdings of those losers) than it sells of all stocks; in quarter 4 this measure rises to 42 percent (t = 1.75). Consistent with window dressing, funds actively sell losers, especially in the fourth quarter.

To check robustness of these results, we have also performed the analysis by value- weighting the funds rather than equally weighting them. This amounts to assuming that all the funds' holdings and trades are done by one aggregate fund. The results, which are not presented, are qualitatively similar, although dumping losers is less pro- nounced when funds are value-weighted. For example, in quarters 1-3, value-weighted sales of losers relative to holdings (corrected for all sales relative to holdings) are 1.0, but for quarter 4, this number is

1.06. This result suggests that window dress- ing is more prevalent in small funds. In- deed, we have performed the analysis for the 20 percent smallest funds as of the first quarter of 1985, and found a stronger propensity to sell losers in general, and in the fourth quarter in particular, for this subsample. For small funds, our measure of selling activity for losers is 1.37 in quarters 1-3 and 1.59 in quarter 4 (t = 2.02). More extensive window dressing by small funds suggests that it may be too hard to fool the sophisticated sponsors of large funds by window dressing, so their managers do not even try.

111. Conclusion

This paper took a preliminary look at portfolio strategies of a sample of 769 pen- sion funds representing about 18 percent of private pension equity holdings in the United States. We have found that the aver- age pension fund is contrarian, in that it buys more intensively stocks that have per- formed poorly and sells a disproportionately high fraction of stocks that have performed well. Such contrarian investment tendencies of the funds suggest that, contrary to a common perception, trading practices of in- stitutions might reduce volatility of stock prices. This result obviously deserves a closer look.

We have also found evidence that in ev- ery quarter, funds sell poorly performing stocks disproportionately to their holdings, that is, they get rid of "mistakes." Moreover, the pace of dumping mistakes acceler- ates in the fourth quarter, consistent with the most common window-dressing strategy. This result is stronger for small funds. This finding supports the view that window dress- ing is a response to costly monitoring by fund sponsors of individual portfolio deci- sions, for it pays less for large funds, whose portfolios are monitored more frequently and completely, to pursue this practice.

Interestingly, purchases of losers also rise in the fourth quarter. This suggests that sponsors do not just look at the funds' hold- ings of winners and losers, but recognize that buying losers after their prices fell might be smart. The overall evidence reveals some window dressing by pension fund managers. However, we stress that fourth-quarter strategies do not radically depart from the usual practice, but rather accentuate the typical investment strategy of funds: buy losers and sell both extreme winners and losers.

REFERENCES

Haugen, R., and Lakonishok, J., The Incredible

January Effect, Homewood: Dow Jones

Irwin, 1988. Jansson, S., "The Fine Art of Window

Dressing," Institutional Investor, Decem-

ber 1983, 139-40. Lakonishok, J. and Smidt, S., "A Seasonal

Anomalies Real: A Ninety Year Perspec-

tive," Review of Financial Studies, Winter

1988, 1,403-26.

What about other forms of window dress- ing? There is weak evidence of increased demand for winners in the fourth quarter: the sale of winners slows down (t = 1.22) although purchases do not rise. In addition, there is strong evidence of an increase in the purchases of losers in the fourth quarter relative to the first three (t = 5.47). In quar- ters 1-3, funds buy 22 percent more losers than they would if they bought stocks ran- domly from the universe, in quarter 4 this number rises to 33 percent. This result is robust in a variety of specifications. It is inconsistent with the simple version of win- dow dressing, according to which just hold- ing a portfolio that performed well (regard- less of how long they have been held) is rewarded by the sponsors, but consistent with the more sophisticated view that buy- ing losers after they have fallen can be justified to the sponsors. Funds might buy losers in the fourth quarter to provide li- quidity for individuals who are selling them to realize losses in the current tax year. In this case, losers might be particularly good bargains in the fourth quarter.

The evidence on purchases and sales re- veals that trading is not always in propor- tion to holdings. Extreme losers represent 6 percent of both buying and selling, but just 5 percent of the holdings. Funds clearly trade extreme losers more. Perhaps they get rid of "mistakes," consistent with window dressing, but they also buy stocks they think are undervalued. Winners in contrast are overrepresented in sales relative to holdings and underrepresented in purchases relative to holdings. One way to summarize these data is that funds are generally contrarian (buy losers and sell winners) except they get rid of mistakes. The next section looks at the data more closely by comparing the fourth quarter to the first three.

Comments
  • Recommend Us