Hi Valeryn, Thanks for replying, so in regards to 3. I think this is a very important subject, and especially difficult when dealing with MR systems and its unique character. As i understand it, you try to keep max dd PER strategy at around 10-12%, and then use around 3-4 strategies, and you look at you max dd as then 3-4 x the 10-12%.. but you would not expect them to hit same time due to your selection or putting together non correlating systems... So in reality you size positions according to % of total portfolio size, but you cap max positions at same time per strategy level. And you look at max DD as strategies combined for example 4 strategies x 10%dd each = 40% DD at portfolio.. And you cap max positions per strategy depending on how much total margin you are willing to take on, at total portfolio level since you are betting 5-10% per position of whole portfolio. This is very simplified, but have i understood your correctly ?
When it comes to DD and return, if correlation is sized up correctly then your return to DD ratio should improve. Typically increased return and ~same combined DD or ~same return and reduced combined DD. If trading stocks, we should be expecting majority of longs to suffer during volatility spike so a portfolio of long strategies can have combined DD increased (this might not show in backtests but you should expect that). Combining short and long strategies has the best potential to reduce combined DD as well as return. No guarantees of course. When markets are relatively calm even a portfolio of long strategies can be uncorrelated. Here is an example. This account only trades 3 long strategies and yet had DD reduced and returns increased. Unlikely to happen if market took a 30% dive in Sept. A probability of that is not 0 but is low enough to be acceptable for me. I am not sure how to quantify it though. Ultimately a trader has to accept a reasonable risk, and we all decide for ourselves what's reasonable. For me it is when each strategy has 10k+ trades, survived multiple market types/cycles/volatility events (I test over 30-70 years), and a combination of them has the desired effect of increasing risk adjusted return. What I do expect to happen - MaxDD per strategy will hit 2x of its' historical max at some point. That might not happen for a few years, but this is practically a guarantee.
Hi Val, To generate the 10k+ trades do you mean that you are testing each strategy over 70 years? or do you want a strategy to generate that many trades over say the last 15-20 years. Then test on previous out of sample data going back that far? Thanks
Sorry for delay, just got back from my vacation. The second one, typically I will have 10k+ trades in backtest over 15-20 years. And I split data into 5 years internals, so in that sense - having 2-5k per each 5 years is quite comfortable for me to make conclusions. And I'll need to see many of those having decent stats. Because I still want to to have a lot of trades over "recent" data to make results statistically significant. For as far as 70 years back, sometimes I'd have more than 100k traders per strategy. But I don't routinely go back that far. Market has changed quite a bit and lets say back then it wouldn't possible to use my edge, but doing so gives me a good sense of what happened during as many volatility spikes as possible and as well as outlier events. Then I'd also look at Monte Carlo simulations, but not very frequently. All of those are pieces of the puzzle and there are many more.
I personally haven't as I don't have good data (or time to find it) for them and it seems that US provides good diversity over companies locations and industries, as well as great liquidity. But other traders did. There are quite a few I know who are trading Australian market, as well as some in Japan/Hong Kong markets, ~same method. The approach I'm using for system development and validation is not market dependant, I just don't like to go blind into a new market with 100% same systems without runnings my tests over relevant data set. Looking forward to Norgate adding Canadian data. I'd certainly try my systems there. Val
did you backtest your system in bear market in the end of 2008 and March of 2020? What is the performance during that time?
I was trading it live thru whole 2020, the very first post in this journal contains stats. With more updates posted overtime. Below is my equity vs SPY from the first post, I was up ~54% by that point. Not only I backtested thru 2008 but early 2000s, late 90-ies and 1987 as well. I don't remember what stats exactly were except that - satisfactory. But generally speaking my systems do better when volatility spikes so those all were likely unusually high return years. To my surprise, backtest vs live came pretty close in 2020. With live slightly over-performing.
Why do people who were not alive in 1987 backtest to 1987? 1987 is not 2021. All everyone does is backtest, backtest, backtest. Time for you to forward test, and draw your conclusions from the future, and not the past.