I'm curious as to how many academic studies and industry white papers are actually using daily data to report intramonth drawdowns; specifically, when the papers are often reporting monthly signals, statistics, and performance. I would think it would be obvious that honest reporting would report risk statistics using daily data granularity, but considering they do not often explicitly state such, it is hard to say with certainty.

Many of the daily data series themselves are hard to gather and guarantee same results (dividends, etc) over the long term. But there are numerous papers on topics such as multi-asset momentum strategies going back to the seventies. There would be a large difference in some risk metrics like drawdown if only monthly closing data points were sampled.

Any first-hand experience or references on the matter are appreciated.

edit: Thanks for replies so far. Just for clarification; I'm not really asking about the merits or pitfalls of sampling at different intervals-- I'm well aware of that. I'm asking more about experiences with various papers(academic) and white-papers(industry) that show monthly statistics back to the seventies (or more) and whether or not you've found that they divulge risk metrics (esp. drawdowns) based on daily or only monthly granularity. It's important for comparison purposes to understand if they are underestimating risk measures in such old data. If some paper, displaying only monthly results, charts, and tables, tells me that the worst drawdown over 40 + years was -25% (some use data going back to twenties), I want to know if that included daily granularity or not. Unfortunately, I don't often see that clarification and so I'm wondering if it is the norm to only use monthly sampling for long term systematic studies with potentially sparse daily data available on total return series. There are some high low data available from CRSP and IDSI going back to the 60s, so I agree with Freddy that it can be done, just more interested in what has actually been applied in papers with older data, so they can be compared reliably.

  • $\begingroup$ just like to point out the problem you describe isn't specific to monthly drawdowns $\endgroup$ – pyCthon Jan 25 '13 at 3:05
  • 1
    $\begingroup$ I do not know how many studies there are but I can tell you that using daily data for the purpose of calculating max drawdowns is the absolute largest granularity I would accept to even look at for strategies of holding periods <1 year. There is just so much half-hearted stuff out there done by people who have never even traded a single stock in their life or understand how an fx roll over is conducted yet they write all those fancy finance papers. I really take issue with from-reality-removed abstraction academic research which is subsequently called financial modeling. $\endgroup$ – Matt Jan 25 '13 at 6:04
  • 1
    $\begingroup$ ...so in this case do not accept drawdown calculations based on monthly closing data points. I have no issue with monthly high-low data point incorporation as long as reasonable fill/stop assumptions are made (or as long as its explicitly mentioned that no transaction related costs are considered). $\endgroup$ – Matt Jan 25 '13 at 6:07

You're right to recognize that the sampling interval of your risk metrics could make a huge difference. For drawdown, there's no reason not to use the same granularity as the data that one has, unless the author is deliberately shaping the curve to appear better than it actually is. A trivial example of this would be if you had a buy-and-hold portfolio in cash equities through May 6, 2010.

However, for Sharpe ratio, VaR etc., it is difficult to get a meaningful value at high resolution.

  • 1
    $\begingroup$ there would be no issue, no matter what granularity used if high/lows were applied and not closing values of compressed data. $\endgroup$ – Matt Jan 25 '13 at 6:47
  • $\begingroup$ I was thinking in the context of intraday strategies. Using highs and lows requires the assumption that the executions are MOC or MOO. $\endgroup$ – elleciel Jan 25 '13 at 9:10
  • $\begingroup$ Well OP made it pretty clear he is not talking about intraday strategies. Those are strategies with daily if not weekly/monthly holding periods. And for highs/lows you need nothing else than the max price and min price. Nothing else. Let's not split hairs here unless the big picture is in place and set right. $\endgroup$ – Matt Jan 25 '13 at 9:11
  • $\begingroup$ Yes, using highs and lows is half-hearted but I understand that you might have your own interests. Since few strategies are MOO/MOC, and maximum drawdown on each single day is not path-independent, you should use the highest resolution of data that is available to you, even if the strategies have holding periods measured in days. $\endgroup$ – elleciel Jan 25 '13 at 9:35
  • $\begingroup$ Thats not what I said. Lets say you only have monthly resolution data available. What is half-hearted in using the min and max in a given month as a clear indication of where prices traveled to at the boundaries? And I do not have any own interests. Please let's stick to the topic at hand. As I repeatedly said you concern yourself with micro execution logic and whether a strategy executes MOC or not while you have not sorted out the 32,000 feet above the ground big picture. (And yes do not test MOC on monthly data, its a dumb approach to start with, so no need to make such assumptions). $\endgroup$ – Matt Jan 25 '13 at 9:47

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.