عملکرد تایمرهای بازار حرفه ای: شواهد روزانه از استراتژی های اجرا شده

We examine the performance of 30 professional market timers during 1986–1994. Prior studies have analyzed implicit recommendations from mutual fund returns or explicit recommendations from newsletters. We analyze explicit recommendations executed in customer accounts. Using four tests, three benchmark portfolios, and daily data, we find significant unconditional and conditional ability that is robust with respect to transaction costs and survivorship bias. Relative ability persists and varies with the frequency of recommendation changes. When recommendations of successful timers are observed monthly instead of daily, significant ability generally disappears. Hence, the frequency with which recommendations are observed can change inferences regarding ability.

This paper investigates the performance of 30 professional market timers during 1986–1994. A notable feature of this analysis is its relatively unique data. Previous studies typically have used implicit recommendations estimated from mutual fund returns or explicit recommendations obtained from investment newsletters. This study uses explicit recommendations executed in customer accounts. These timers have limited power of attorney, indicating that their clients trust their ability. They disclose their recommendations voluntarily, suggesting self-confidence in their performance. If any timers possess significant ability, the timers studied here are likely candidates. We use four tests to analyze performance on a daily basis: mean-variance tests, unconditional and conditional Cumby and Modest (1987) regression tests, and Graham and Harvey (1996) weight change tests. Each test uses benchmark portfolios that correspond to Nasdaq, NYSE/Amex/Nasdaq, and the Standard & Poor's (S&P) 500. We find evidence of significant ability across all tests and portfolios. This evidence is robust with respect to transaction costs and survivorship bias. We also find that relative performance persists and varies with the frequency at which a timer changes recommendations. For comparison purposes, we examine a consensus timer, whose recommendations reflect forecasts based on all 30 timers, and a statistical timer, whose recommendations reflect forecasts based on standard macroeconomic variables. The consensus timer demonstrates both unconditional and conditional ability. Because the statistical timer forecasts based on conditioning variables, we expect it to show only unconditional ability, and that is the case. Finally, we obtain an interesting new result involving the frequency with which recommendations are observed. When we reconsider our most successful timers, but update their recommendations monthly instead of daily as done originally, we find that their significant ability vanishes. Thus, inferences regarding timing ability can vary dramatically depending on how frequently recommendations are observed. The paper proceeds as follows. Section 2 discusses prior research and methodological issues. Section 3 describes the data, and Section 4 presents initial results from various market timing tests. Section 5 examines timing ability in more detail, exploring topics such as consensus and statistical forecasting, survivorship bias, persistence, transaction costs and management fees, the frequency with which a timer changes recommendations, and the frequency with which a researcher observes recommendations. Section 6 offers concluding remarks.

We provide new evidence regarding the ability of professional market timers. A notable feature of the analysis is its relatively unique data; i.e., timer recommendations that are explicitly known and executed in customer accounts. Using four types of tests and three benchmark portfolios, we examine both unconditional and conditional timing ability on a daily basis. Contrary to most prior research, we find evidence of significant ability across all tests and portfolios. Transaction costs and survivorship bias reduce, but do not eliminate, evidence of significant ability. Relative performance persists over time and varies with the frequency with which a timer changes recommendations. We construct hypothetical consensus and statistical timers and evaluate their performance. The consensus timer displays both conditional and unconditional ability. Because the statistical timer forecasts based on certain macroeconomic variables, we expect it to show only unconditional ability, and that is the case. When recommendations of successful timers are observed monthly instead of daily, significant ability generally disappears. Thus, the frequency with which recommendations are observed can seriously affect inferences regarding timing ability. Although we find evidence of ability across all portfolios, the evidence generally appears strongest relative to Nasdaq and weakest relative to the S&P 500. We can only conjecture as to why this pattern occurs. We know that these timers usually invest in small capitalization or growth funds. If timers provide recommendations tailored for a fund that is more similar to Nasdaq than the S&P 500 in terms of its composition, then it is not surprising that the recommendations fare better relative to Nasdaq than the S&P 500. For example, certain timers might use a forecasting model that incorporates the relatively high autocorrelation exhibited by Nasdaq returns, in which case their performance should look better relative to Nasdaq than the S&P 500. However, the apparently better performance obtained using Nasdaq returns might be spurious, in the sense that it could reflect nonsynchronous trading and nonexploitable return autocorrelation. In closing, we believe three factors potentially explain why this study, unlike most of its predecessors, finds evidence of timing ability. First, the timers studied here are professionals who execute their recommendations for clients. If any timers possess significant ability, these timers are likely candidates. Second, because the recommendations studied here are explicitly known, they are free of estimation error, unlike implicit recommendations estimated from mutual fund returns. Third, the recommendations studied here are observed daily, not monthly or quarterly. Daily observation could aid detection of significant ability, especially for timers who frequently switch between asset classes. However, nonsynchronous trading does not explain our finding of timing ability. We know that timing ability appears weakest for the S&P 500, and for the S&P 500 nonsynchronous trading does not lead to spurious evidence of ability. Consequently, nonsynchronous trading cannot explain our conclusion that significant timing ability exists.