Inside the Black Box: The Simple Truth About Quantitative Trading
The book makes a mess of the distinction between Alpha, which is earned from other active traders, and Beta, which is earned from buy-and-hold investors. What he calls "theory" in a strategy is no more than ad hoc marketing junk. Theory does not mean just saying you exploit a "documented behavioral bias" or "institutional rigidity." It means a real, sensible, testable theory of who is losing the money you're making. You need to know who those people are, why they are doing it and monitor that they keep doing it. Without a theory the only way you know your strategy stopped working is when you lose money, you never have warning, and you never know when it's safe to go back to it. Also, a theory tells you what to do when things stop working, the author seems to suggest that your only options are keep the strategy running, change it or shut it down. Professionals have several layers of backup plans. Theory is what separates a quant trader from a technical analyst.
Risk management is covered only in the portfolio management sense, in which risk a constraint or something to be minimized. Independent risk management is barely mentioned, and completely misdescribed. The author does not know what Value-at-Risk is, any paragraph with that term should be ignored. The first thing to ask any quant trader for is her VaR backtest. She should produce a number every day before trading starts such that she loses more than that amount 1 day in 20. The backtest should show the right number of break days, subject to statistical error, and those breaks should be independent in time and of the level of VaR.
Risk management is covered only in the portfolio management sense, in which risk a constraint or something to be minimized. Independent risk management is barely mentioned, and completely misdescribed. The author does not know what Value-at-Risk is, any paragraph with that term should be ignored. The first thing to ask any quant trader for is her VaR backtest. She should produce a number every day before trading starts such that she loses more than that amount 1 day in 20. The backtest should show the right number of break days, subject to statistical error, and those breaks should be independent in time and of the level of VaR.
If you can't produce a good VaR, you don't understand your everyday risk, what happens 19 days out of 20 when markets are normal, so you can't possibly understand your tail risk. VaR is not a measure of risk, it tells you the range in which you can trust your models. You worry more when it is too small, when your models can only be validated in narrow circumstances, than when it is too big. It's not that you like losing money, but for two strategies with the same return and volatility track record, you trust the one that has survived significant adversity more than the one that has seen only mild days.
I think I've just told you everything bad about this book. Note that it's less than 1% of the length of the book. That pretty much sums up my judgment, this is a great book, 99% pure.
No comments:
Post a Comment