I got a question, which i don't have an answer to, because it would require me to get a working strategy in the first place to test this out, but I think this is a good idea, so I count on your experience to help me So, if we have a robot that trades in multipairs, the drawdown is the enemy right, but I thought of a method to limit DD : the idea is : let's set a drawdown limit for each pair : then, do we have a DD higher than this limit on one pair ? - If yes, then do not open other trades on other pairs, maybe even close the other trades, and wait for the DD to get back to normal. - If not, then open orders on other pairs as usual. This way can we limit the max DD? If the set threshold is 2 percent for instance, then we stop trading on the other pairs (that could add to the max DD) and we wait to the DD to be resolved on the culprit pair. I know that trading on multiple pairs lowers the DD globally, because max DD of every pair is rarely occuring at the same time on every pair, but what about the idea to stop trading on other pairs if the DD of one pair is higher than a threshold ? thanks for your participation Jeff
What do you mean by Drawdown ? Unrealized loss ? Usually a drawdown is the difference between your all time equity (bankroll) high and your current one. Max DD is the greatest high to low ever experienced. The only way I see to diminish variance is to lower exposure. Diversification lower the overall variance of your portfolio only if you invest into un-correlated assets. Drawdown is not a problem if you allocate no more than what ought to be risked. DD is only luck around your edge ... You can’t control luck. Only your exposure.
yeah i get that if you limit exposure (i.e trade less often) you'll lower the max DD. the more you trade, the higher the max DD will be. But at one moment you will continue to trade and trade more and more, so I don't think limiting trading is the solution. here is how I see my idea : you can compute a max DD for every pair you trade (let's say 28 pairs) in the backtest there's the risk (small but not inevitable) that the max DD occurs right at the same time on all 28 pairs, which would make a very big DD. So my idea is to stop adding positions, when a certain max DD is reached on ONE pair. that way you limit (or even cancel) the risk that all 28 pairs are at max DD at the same time I also get the uncorrelated thing, it's useless to trade correlated pairs, but I think it's not that important, as soon as you limit the number of simultaneous pairs trading at the same time during DD peaks. Also correlation shows how pairs behaved in the past. Things can change in a second. Hard to predict correlation relationships, even cointegration Jeff
By reducing exposure I meant, reducing the size of your overall positions. As you can’t reduce the probability of unfortunate events but you can reduce its impact by lowering exposure (Bet size). Max DD isn’t a fixed number. You believe -10 is Max DD but your portfolio can be down -150 with 10 pairs. What do you mean by “Stop adding position” ? Adding size to the 28 pairs ?
One example: Portfolio composed of two un-correlated instruments. Potential returns for each {-1,1} You can experience 4 situations at time t 1. {-1,-1} : -2 2. {-1,1} : 0 3. {1,-1} : 0 4. {1,1} : 2 You can’t reduce the probability of #1 happening, But you can limit its impact by betting smaller.
when i said "stop adding positions", i meant stop opening new orders on other pairs. if you have potential returns {-1;1} on 4 pairs We know that the drawdown max of any pair was -1.0 over 20 years of 28 pairs backtest. now let's look at two possibilities: First we trade all the 4 pairs all the time: t=0 pair1 : 0 pair2 : +0.1 pair3 : -0.2 pair4 : 0 maxDD = -0.1 t=1 pair1 : 0 pair2 : +0.1 pair3 : -1.0 pair4 : 0 maxDD = -0.9 t=2 pair1 : -1.0 pair2 : -1.0 pair3 : -1.0 pair4 : -1.0 maxDD = -4.0 Now with my method, if we set the threshold at -0.5, we then stop opening trades on other pairs than pair3 from t=1 (because pair3 DD = -1.0) t=2 pair1 : not trading, only closing pair2 : not trading, only closing pair3 : -1.0 pair4 : not trading, only closing maxDD = -1.0 t=3 pair1 : not trading, only closing pair2 : not trading, only closing pair3 : -0.1 pair4 : not trading, only closing maxDD = -1.0 we are now above the treshold (-0.5) so we can start opening new trades on other pairs. the Max DD was -1.0, compared to -4.0 with your method. Of course the max DD on a single pair can be higher than in the backtests, but we limit the probabilities of a big DD based on enough backtest data. That's all we can do The drawdown has here been divided by 4.0, and the profits only a little bit impacted And of course the more pairs you trade, the higher the drawdown reduction this example is a bit simplistic, but please consider only the main idea Jeff