Steve Mayo Legend  Posts: 414 Joined: 10/11/2012 Location: Austin, TX  User Profile | In that tool, I calculate a frequency distribution with box plot on the expected returns based on a Monte Carlo simulation that uses "bootstrapping" to compensate for exactly that issue. In other words, it takes thousands of random samples of "possible" returns based on historic daily performance and shows you the probability of achieving a certain annual return using parametric and non-parametric techniques. Basically, as Steve2 points out, the backtest is only ONE possible outcome, with nothing to tell you how likely THAT particular outcome might be. You really can't make a statistically-sound judgement based on that single equity curve, or those single-number metrics like CAR and MDD without knowing the probability range, specifically mean and standard deviation, or median and interquartile range. Even with this tool, I would still caution you to do what Steve2 suggests as well. Rerun your simulation with differing start dates, differing "Sort By's, and different symbol lists to get a better feel for the likely performance. Also, it's best to allocate "blocks" of funds to each strategy so that one strategy doesn't compete so much with another, and then keep the allocation/trade low so that your strategies only rarely ever hit the equity ceiling. | |
