A few years back, I had a great discussion with die-hard Trend Follower, Scot Billington. I would like to share some key moments with you, including how to perceive past track records, and also some of the differences between the long side and the short side. If you would like to listen to the full conversation, just go to Top Traders Episode 25.
Deciding between Systematic and Discretionary Trading
Scot: ...I started putting together a trading model. In my opinion, they were three big picture decisions that somebody had to make when you talk about what kind of trader you are going to be. The first was discretionary or systematic and mechanical.
So I would define discretionary as I bring in different inputs, whatever those inputs might be. I weigh them in a non-standard fashion, meaning that I don't weigh them necessarily the same way every time. I might bring in the same inputs, I might look at different ratios, but sometimes input A overwhelms input B, and sometimes input B might overwhelm input A. Regardless of what that might be, and then I would make the trade decisions in that fashion. Systematic I define as I do the exact same thing every time. I might argue that, if you have any discretion, then you are discretionary. So that even if I have a mechanical model, but I decide 7 times a year to override it, I suspect those seven times a year are going to be 7 of the more volatile and the larger outcome periods, and in essence you have a discretionary model, which is fine but you are a discretionary trader.
"The primary reason that we wanted to be systematic is that we felt like we wanted the emotions taken out of the trading process."
The reason that I and we have gone with systematic is three-fold. The first is that we wanted something that the efficacy of which could be at least estimated through historical modeling and backtesting and the like. So if I'm a discretionary trader, one of the difficulties that we found was how do I know that my theory is accurate? I think that XYZ, whatever XYZ is, it might make perfect sense, but what I have there is a good hypothesis that will be interesting to test, but I really have no method of testing it. Therefore, no way to prove that at least my idea had worked in the past. The second reason that we went with a more systematic model is that we felt that it would be much easier to apply to a wide variety of markets. If I was going to be..and that also ties into the inputs we might use...but if I were going to be discretionary, and I was attempting to trade the Yen and cotton, it would perhaps be very difficult to be an expert in both of those two markets. Now you could be discretionary and not necessarily have fundamental inputs, but that will be the second part of this answer. The primary reason that we wanted to be systematic is that we felt like we wanted the emotions taken out of the trading process. We wanted something that was repeatable, so that I could say the same decisions that I made in June of 2004 I'm going to make in December of 2018. It's the same process. It's a much more repeatable process than my weighing all these different factors. The other thing is that we think that it's...I don't know about impossible, but extraordinarily difficult to separate your decision-making process from your own emotional state at any given time. I think it's probably a bit fanciful to say that I would make the exact same decisions on the day that my wife left me as the day that my son won an Olympic gold medal. (This applies) also within trading: If I've just had 4 straight up 15% months I think it's extremely difficult to bring the same analysis as if I just lost...if I'm in the middle of a 30% drawdown. What we basically said was that my wife has just walked out on me should not have any effects on the trades that I take.
So the second thing that we looked at was what kind of timeframe are we going to trade? If you look at timeframe as a spectrum, am I going to be shorter term or longer term? Is there some logical reason to make that selection? We've elected to be longer term for one, and it's not a very sexy reason, but it's in our minds it's extremely important, is that random trading's expected outcome in a frictionless or costless world would be to break even.
So if I'm trading randomly, my expected loss is my cost. My commissions that I pay, and more importantly the bid/ask spreads that I pay. That's also going to encompass any kind of gaps or slippage in a fast market, but again that's just a wider bid/ask spread. Those costs come into play every single time I trade, but they're fixed, so if I hold a trade for 8 minutes, the bid/ask spread is just as wide as if I held it for 8 months. I might get a little break on the commission from my brokerage, but at the end of the day, the holding period of my costs has no impact on my costs. My costs are going to be fixed regardless of my holding period. So if you look at it from that perspective and you say that the amount of what we call a trading edge is a non-random entry and exit decision. It means that I have some non-random method of deciding when I'm going to buy and when I'm going to sell. The amount above randomness that my method needs to have to break even is my costs. The costs are, in casino terms, the cost of the house edge. That's how positive I need the deck to be, or that's the amount that I need to be able to forecast future price moves to break even. So we would consider forecasting future price moves to be extraordinarily difficult. Therefore, we want that hurdle to be as little as possible. Does that make sense Niels? Do you see where I'm going with that?
Scot: Now, there is another side of that coin. A shorter term method is going to have more instances in a given time period, and therefore a smaller net profit, meaning after cost profit can be profitable, or can have a good method. So I need a larger gross edge, because I'm going to give up so much more of my edge in the costs, but a smaller net edge can be profitable in a shorter term. Does that make sense? So with that, the lower hurdle to clear with the lower cost parameters, to us was an overwhelming argument for having a longer term method.
Niels: Absolutely, but since we are talking about costs at this point, I'm just wondering...I mean obviously cost today for trading is a lot lower than they were in the mid-1990s, does that change your view in some sort?
"..At the end of the day, the future might be completely different to the past."
Scot: Some, but not particularly. It's still...we track our costs very closely. When you add it all up, and average them over 30,000 contracts we traded in X amount of years, we still see, including roles, we still see about $35 a round turn. So if I'm a CTA that does 3,000 round turns per million, 35 times 3,000, that's $105,000 a year, that's 11%, so the best guys in the world make 20%, that's half of that. So if I just do immediately 1,500 round turns, this other guy's got to beat me by 5 1/2% a year just to tie me. You see what I mean? We'll probably touch on this latter, but to us that's a fact that's not an opinion and it's something that's based on empirical evidence. That's just a straight mathematical fact. So when we get into the modeling and testing, and all that, there's a huge weakness in that it's all empirically based, and I can try to make it my empirical base as solid as possible. I can try to make it as robust as possible, but at the end of the day, the future might just be completely different than the past. In which case all of those...that's the black swan kind of idea that Taleb put forth in his series of books, and it's a very accurate one. However, if I'm saving 5% a year in costs, that's not perceptible to a black swan. In fact, it can only be helped in that the less I trade, and the wider my trading parameters are, the less impact massive gaps would have on my outcome. You see where I am there? So what I want to do is I want to line all of these facts up in my favor before I'm forced to use empirical evidence. Does that make sense?
Niels: Sure, sure.
Scot: Think about it this way, a truly losing trading strategy, by definition, has to be as rare as a truly winning one. Right? Well, I could just take the opposite of the trades. If you had a truly negative losing expectancy, that is very valuable, because I could just take the opposite of your trades, and I would make money. So those have to be very rare, correct? Or as rare as a winner - which means that most trading is random. People think that factor A, B, C says something about the future of price movement, but that factor is either fairly valued at the current price or does not have any impact, and it's no different than my drawing a trade out of a hat. But just because I draw a trade out of a hat, that doesn't mean that trade's going to be a loser, it's just a random trade. So most people are trading randomly with an expected loss of cost. Almost by definition, that almost cannot not be true. So I all of these people and I say, OK, well Janet Yeltsin is going to say this thing about whatever interest rate, and OK if that happens the dollar is going to do Y. When they go through there they might sound very smart, and I don't doubt they're well educated, and it might be a well thought out opinion, but by definition that is either not predictive of the future market move, or it's already been fairly valued by the market, because most trades that people put on have to be random. They're not losing, their random. You see what I mean? I was a market maker, and I would stand in my pit and I would look around and maybe not including myself, but I'd think...because a lot of guys made a lot of money, I'd be like, you know, there are 100 guys in here and the average take-home after their own costs of paying commissions and paying clerks, renting the seats, and all that is maybe 1/2 a million dollars, so that's 50 million dollars a year that this pit makes. Well, who pays that? It's the people who want to take positions. Right?
Scot: So when we look at that, and we think about costs, we think about this is exactly the amount of nonrandom price behavior I have to capture to break even and then go on to be profitable. The lower that cost is, the less of anomaly I have to capture, and, even more importantly, the more room I have for the future to be worse than the past.