Can hedge funds’ characteristics be exploited to pick hedge funds? Is a characteristics-based strategy more profitable than a naïve strategy? Joenväärä and Kahra address these questions by using three hedge fund characteristics—managerial incentives, the length of the notice period, and fund size. This approach is derived from a previous paper by Brandt, Santa-Clara, and Valkanov (2008) {1}, who exploited the characteristics of equities to build an optimal equity portfolio. The authors’ work assumes that these characteristics impact hedge fund performance, as looked into by other studies that focus on explicit micro-factor models.

The period they analyse goes from January 1995 to December 2006, and data are provided by Hedge Fund Research. After correcting for the backfilling and the self-selection biases, the database contains 2,824 individual hedge funds; they are active in one of five strategies: directional trading, relative value, security selection, multi-process, and managed futures. Managerial incentives are measured by the manager’s option delta, which is “the sensitivity of the manager’s compensation to a one percent increase in the fund’s net asset value.”

Over the entire period, the estimates of parameters representing hedge fund characteristics show that the three parameters are jointly statistically highly significant. The size parameter has a negative coefficient, while the notice period and the managerial incentives have positive coefficients. In other words, it would be better to invest in a small fund with high managerial incentives and a long notice period than in a big fund with low managerial incentives and a short notice period. On the basis of their characteristics, weights are attributed to the analysed hedge funds. Each year, they are ranked by weight and sorted into quintiles. Each quintile forms an equally weighted buy-and-hold portfolio. Three performance measures are then calculated: alpha, the information ratio, and the Sharpe ratio. Alpha is obtained through Fung and Hsieh’s seven-factor model, in which the factors are the excess return on the S&P 500 index, the spread between the Wilshire small cap and large cap returns, the change in the ten-year Treasury yields, the yield spread between the ten-year Treasury bonds and Moody’s Baa bonds, and the excess returns on portfolios of look-back straddles on the bonds, on the commodities, and on the currencies. The information ratio is the ratio of this alpha to the standard deviation of monthly alphas.

This process is applied successively to the entire period, which is called the in-sample period, to the period from January 2002 to December 2006, called the out-of-sample period, and to the out-of-sample period with only feasible strategies. To build feasible strategies, the following funds are excluded: funds that have a lockup restriction, funds that do not provide redemptions and subscriptions at the end of the year, and funds whose notice period is longer than 120 days.

For the in-sample period, the seven-factor model displays an adjusted R² ranging from 66.3% to 73.8% across portfolios. Regardless of the performance measure, the spread between the top and bottom portfolios is statistically significant. The top quintile portfolio posts a Sharpe ratio of 0.441 and alpha of 0.07, while the bottom portfolio posts a Sharpe ratio of 0.282 and alpha of 0.028. For the out-of-sample period, the seven-factor model exhibits an adjusted R² ranging from 61.5% to 77.6% across portfolios. The top quintile portfolio posts a Sharpe ratio of 0.562 and alpha of 0.06, while the bottom portfolio posts a Sharpe ratio of 0.427 and alpha of 0.033. For feasible strategies over the out-of-sample period, the seven-factor model exhibits an adjusted R² ranging from 47% to 64.3% across portfolios. The top quintile portfolio posts a Sharpe ratio of 0.657 and alpha of 0.079, while the bottom portfolio posts a Sharpe ratio of 0.472 and alpha of 0.049. Regardless of the performance measure and the sample, the top quintile outperforms the bottom quintile.

To compare characteristics-based and naïve strategies, for each sample (in-sample, out-of-sample, out-of-sample with a feasible strategy), funds are sorted into twenty-five portfolios. In the first approach, funds are sorted into quintiles by the t-statistic of the seven-factor-model alpha, and each quintile is divided into quintiles by weight. The second approach consists of sorting funds into quintiles by fund size, and each quintile is divided into quintiles by weight. The third approach consists of sorting funds into quintiles by manager delta, and each quintile is divided into quintiles by weight. The fourth approach sorts funds into quintiles by length of notice period, and each quintile is divided into quintiles by weight. It controls for effects related to fund size, manager delta, and the notice period. Again, spreads between top and bottom portfolios are calculated. The authors conclude that characteristics-based strategies outperform naïve strategies.

{1} Brandt, M. W., P. Santa-Clara, and R. I. Valkanov, (2008), Parametric portfolio policies: Exploiting characteristics in the cross section of equity returns.