I’ve said this over and over that if you are not reading Matt Levine’s free daily newsletter you are really not an informed market actor. The man writes so well about so many complex financial issues that his daily missive is often the highlight of my day.
This week in a riff on Bill Gross and the meaning of Alpha, Matt truly outdid himself and I am going to shamelessly quote a very large piece of his note because I think it carries so many important lessons to those of us who switched to algorithmic trading.
Levine writes, “Did Bill Gross generate alpha? Well, and what if he didn’t? What is “alpha”? Often you read that alpha is an investment manager’s return above a benchmark—if the S&P 500 returns 10 percent and a stock manager returns 12 percent, he has added 2 percentage points of alpha—but academics and allocators tend to take a stricter view. If he just bought riskier stocks to get that extra return, that’s not really alpha; he’s not demonstrating any extra skill or “really” outperforming the market.
One stricter approach goes something like this:
1. Look at the manager’s returns over time, and get a rough sense of what he actually did to get those returns.
2. Construct some smallish number of mechanical investing strategies that are sort of similar to what he actually did. These strategies could be as simple as “buy all the stocks in the S&P 500 index” or as complicated as “use an optimal trend-following strategy of buying lookback straddles”; they could involve a passive buy-and-hold approach or constant trading; but the point is that they can be totally specified in advance and a fairly simple robot could carry them out.
3. See how much of the manager’s actual performance could be explained by those mechanical strategies: That is, if you had just replaced the manager with a handful of simple robots programmed to carry out straightforward strategies, how close would the robots have come to his actual performance?
4. If the robots’ performance looks nothing like the manager’s, then you have just chosen the wrong strategies: If there is little correlation between the mechanical strategies and the manager’s results, then that means that the manager is doing something very different from what the robots are doing, and you have learned nothing.
5. If the robots’ performance looks a lot like the manager’s—if the correlation is high—but the manager outperformed the robots, then he is adding alpha: He has demonstrated skill that your simple robots can’t match. His strategy is not as simple as “buy all the stocks” or “buy all the stocks with high book values” or “buy all the stocks that went up yesterday” or anything else that you can fully describe in a sentence; his strategy instead involves buying stocks that are good and not stocks that are bad, based on his own mystical intuition or hard work or whatever.
6. If the robots’ performance looks a lot like the manager’s, but the robots outperformed him, then he has negative alpha. Perhaps this just means that he’s terrible and keeps losing money, but if you’ve come this far that is unlikely to be the explanation. Instead, what is more likely is that he has mostly made money, and has attracted investors and made a name for himself, but the way that he has made money is not primarily through mystical intuition about what stocks to buy. His intuition about what stocks to buy is mostly bad—worse than the robots’ mechanical selection—but his choice of strategies worked out fine.
Now the money line in this whole long explanation is the very last sentence. “His intuition about what stocks to buy is mostly bad -—but his choice of strategies worked out fine.” Substitute the word currencies for the word stocks and the concept can be applied to any one of us. THIS is the key insight that makes me so excited about algo trading. The beauty of algo trading is that you do not have to make great trades. All you need to do is just make good enough trades -- AS LONG AS YOUR STRATEGY IS THE RIGHT ONE. This now turns you from a trade idea generator to a manager of strategies, which you can then compile into portfolios to make pips something like this.
|
