Schneeweis: Monthly hedge fund data might be leading us astray

14 March 2007

Thomas Schneeweis is an academic. So he views the primary benefits of managed accounts not as operational, but as analytical. He argues that daily and weekly return data (usually only available via a managed account) is superior to the more common monthly data used in the hedge fund industry. Schneeweis tells All About Alpha that monthly data hides all sorts of things from, for example, volatility during the last week of every month, to a high correlation to intra-month shocks. (Ross Miller of SUNY Albany shows the power of daily data in his recent paper on Fidelity Magellan’s uncomfortably high market correlation.)

Since monthly data points are generally in short supply (most hedge funds have been around for less than 100 months), all available data points are usually used to calculate things like beta, volatility and, of course, alpha. But as Schneeweis points out, the composition of the S&P 500 has changed over the past 5 years. So why examine 5 years of hedge fund correlations?

Furthermore, kurtosis, skewedness and even beta can be influenced for years by one or two outlying historic datapoints. More frequent data means more recent data – and with it, a more accurate snapshot of current portfolio positioning. Schneeweis told the gathering:

“If you use monthly data and you count August 1998 in there, it affects your data for the next 3 years (if you’re using 3 years to come up with your beta). Monthly data continues to influence your view on things long after that view is no longer valid.”

He comes armed with research to back up this concern. According to a study conducted by the Center for International Securities and Derivatives Markets (CISDM), daily and monthly return streams for the same fund can yield very different correlation, volatility, skew and kurtosis. For example, the HFRX has a 0.77 correlation with the S&P 500 using monthly data, but only a 0.53 correlation using daily data. He shows similarly dramatic differences between skewedness and kurtosis depending on the types of return streams used. His point is that monthly data is not necessarily wrong, it just needs to be interpreted with a wide confidence interval. Since there are more weekly return data points, the confidence interval will be tighter for weekly return streams.

Schneeweis cautions that while daily data is certainly the most sensitive to current positioning, it might actually contain “too much noise”. As a result of this and other concerns, Schneeweis prefers weekly data to both daily and monthly data. Said Schneeweis:

“I know we love monthly data. But daily data has important implications for how we measure risk. You really do need more frequent data to analyze a manger’s record…I like weekly data the best.”

The implications of this are tremendous. For example, it could easily happen that an investor picks a fund with certain risk characteristics (based on monthly data) for a managed account and then finds that the risk characteristics (based on the weekly data available from the managed account) is actually quite different.