In Part I, I discussed the results of the first incarnation of CASTrader, and the good points of it's performance. There is a lot to be encouraged about, as CASTrader appeared to perform well most of the time of the 1916 to 2006 backtest. CASTrader I is a rousing success, because it succeeded in finding patterns in the market. The times where CASTrader performed poorly are saved for discussion in Part III (the ugly) because the results are truly ugly. This post is not concerned with the bad performance periods, but rather the bad aspects of CASTrader, the back test and why the performance is not as good as it appears (graph at left):
- No friction or transaction costs. CASTrader made some 9000 adjustments to capital allocation over 90 years to achieve the apparent 24% annual performance. This equates to a measly 0.22% gain per trade. Transaction costs in any real scenario would swamp this result if CASTrader actually traded that often - eliminating all the gains and then some. These numbers are misleading, however. Oftentimes, CASTrader made little or almost no adjustment to it's net position, effectively making no trade, especially before 1983. I estimate nearly 3/4 of the "trades" are this type and could be eliminated with no consequence. If so, this raises the profit per net trade to 0.8%, which is much better, but still too low to cover transaction costs (except maybe via the futures market). Unfortunately, this would make the 1915-1983 period look worse, and offset much or all of the benefits of accounting accurately for interest rates. The good news is that from 1982 to 2006, transaction costs are irrelevant, because the returns are so good. CASTrader needs to be taught to trade less when the rewards are smaller, and I'll tackle that in Part IV.
- No intraday data. The simulation is run on daily data simply because that is what is available. Intraday data would make CASTrader behave differently, pure and simple. It's difficult to say performance would be better or worse with intraday data, but for now, I'll assume worse.
- Paper vs. real markets. Everything is usually better on paper than in reality. The absurd bankroll CASTrader ends up with in the end simply becomes unrealistic. You simply can't move that much money in and out of the markets without affecting the very patterns you are attempting to exploit and thereby dissipating alpha. A 90 year performance in one market like that is completely illusory. That said, I have nowhere near that bankroll to play with, so at least starting out, I would expect my impact to be minimal. Nevertheless, despite a small starting bankroll, things just don't work out the way they do on paper - profits will be less.
- Black Swans. The performance from 1983-2006 is encouraging, but even if the diatribe of bad things listed above does not dissipate returns, a Black Swan could. The simple fact is that this period is characterized by the emergence of super traders that appear to have the market figured out, but they are leveraged to the point that an unanticipated event could take them completely out of the game. The paradox is that they seem to perform well when some milder Black Swans like the 1987 crash and the 9/11 occur, as they either were mildly short or mildly long, resulting in good relative performance. In fact, there is little drawdown to be seen anywhere in the equity curve for most of the 1983-2006 period given the apparent astonishing performance. Such conditions are an accident waiting to happen, however. Leverage simply leaves open the possibility of complete ruin, ala LTCM and Amaranth. Although traders are using the Kelly Formula, which in theory precludes the chance of complete ruin, they are using a very liberal version of it that does not fully factor in Black Swans that will eventually occur. I'll discuss ways to solve this problem in Part IV, but it's going to be tough to solve without evaporating large chunks of alpha.
- Tendency to develop zero diversity. The emergence of elite traders means the system evolves towards zero diversity. This isn't exactly what I was hoping for, but I'm now resigned to the fact that any system of traders is eventually going to have wealth distributed according to a power law. A tiny few will allocate most of the capital. I just didn't think it would be this stark. This isn't necessarily bad if the elite are good at what they do - obviously by emerging above the rest, they are, at least until a Black Swan takes them out or conditions change. The problem occurs when transitioning a paper-based system to the real world. I don't want a few elite algorithms that favor leverage to be thrust into an environment where things may be different and then suddenly fail with my hard-earned capital. I'd much rather transition to the real world with full diversity or partial diversity and thereby be less susceptible to losses. CASTrader must start with more full or partial diversity.
The above points are enough to dampen enthusiasm for the substantial good points of CASTrader, but they pale in comparison to the truly ugly I'll discuss in Part III. I'll discuss some ways in the good may triumph over the bad and the ugly in Part IV.
Comments