Anyway. When you are ready to share your track record let me know. I don’t need to know too much about what the code or algo is. I care more about the Sharpe Ratio.
Will keep you in mind! One thing I’ve always wondered: why do you care about sharpe? Not calling you out specifically, im using “you” generically. Like if I could offer a CAGR of say 20% and a drawdown around 20% but a sharpe of 1, wouldn’t that be better than a cagr of 10% drawdown of 10%with sharpe 2 (assuming neither use leverage)?
Sharpe ratio is the gold standard for this industry as long as you are not selling options. Option sellers naturally have high SR. There are hedge funds with returns less than 10% with a 2.5 SR and they have billions waiting to get in. I worked for one of them. Investors like to sleep well knowing that their capital is safe.
If your returns are good you can invest for me and I can pay you 20% of returns. Not much legality if you put the orders in my account through the API. This is called a managed account.
Thanks for the post, and congrats on the great results.
I have a question regarding higher time-frame strategies (daily/weekly/monthly). I’ve noticed that many traders (including you) successfully run systems on these time frames, but I’m trying to better understand the backtesting and validation process.
When trading on higher time frames, a 10–20 year backtest often results in only 100–300 trades (aka statistically insignificant). How do you evaluate and gain confidence in a strategy with such a limited sample size?
Do you rely less on statistical significance in these cases and more on structural or economic reasoning behind the strategy? I’d be very interested to hear your perspective 🙏
Actually most of my long term system have 800 to 1,000+ trades believe it or not. But not all. I generally test into the 1990s so get a lot more data.
Generally I’ll run a lot of robustness testing to ensure the system is actually measuring the signal and not curve fit to few trades. I’ll force the system to swap out the trades taken in the Backtest with other valid signals generated by the system but skipped due to position number constraints. Like a resample Monte Carlo.
Thanks for the reply. So, do I understand correctly that you aim for at least a 1,000-trade sample size, and if a strategy has fewer than lets say 300-500 trades, you try to increase the sample size? What is the minimum required number of trades for you to consider a strategy statistically reliable?
Yes I try to increase sample size either by letting the system hold more positions or forcing different trades but still viable signals into the system (think testing the system on a similar but different universe or forcing the system to skip a trade and take a similar but different trade).
More trades is always better but I try not to get hung up on something having too few trades. If the system is exploiting some phenomena that I have reason to believe is real (momentum for example) I may be okay with a smaller sample. I’ll just be sure to keep the strategy super simple. For example I have a seasonality system that has ~240 trades in it. But the seasonality measurement is super strong and persistent. I still trade it even tho it has a small sample size. I just make the system super simple with 2 parameters.
Is correct that 300 trades are not so “many”, but you have to consider that in 10-20 years, the market faced every scenario, bearish, bullish, sideways, so if the “system” created 300 successfull trades in 20 years with positive result, I consider it a valid system
Hey 👋🏻 No worries, thanks for your opinion! I agree it depends on context and logic of the strategy (That’s why I initiated this thread😅)
like as Author said, if this is strong seasonal tendency - we might be fine with smaller sample size (because there is some fundamental logic behind the price movement)
However, I would not consider 200-300 trades sample size for EMA 21 crossover on weekly timeframe (on some random market), since this is looks like overfitting + low sample size.
I do the same with my Mean Reversion strategies. I’m not a fan of Mean Reversion either, multiple strategies spread out the suck.
You run a lot more positions than I do, I use 1 position size formula for all strategies. Max allocation percent vs max ATR volatility. Some strats use stops but fill-to-stop risk is never ask issue with small sizes. On strategies I want more allocation to, I’ll add a x2, or whatever, to the quantity formula in the strategy.
I would like to see your correlations, both returns and dds, for your portfolio. I’m about to go over mine for the year and reduce or cut the highly correlated.
Mean reversion can suck at times. Spreading out the suck is key! I’ll have to look at the correlations this year. I can see visually on the pnl plots there were some days where correlations were 1. Best way to make that smoother is to add things completely different than what I currently have
Thanks for sharing. How do you size your positions in your strategies? Is this fixed percentage, vol target or anything else? Do you have an upper % limit for single position and ticker?
It’s a mix of equal weight and vol based. I don’t have an upper limit. If 3 systems say buy xyz ticker I buy it with essentially 3x size. If 3 systems say long that means the signal is strong so I accept the concentration risk. I’m holding 140 other positions anyway. Most of what I do is single stocks. I have some lev ETFs but thats a relatively newer addition.
Thanks. So multiple strategies might enter trades for the same ticker at the same time so there's no limit on that. I guess you do have some limit on each single position for a specific strategy? From your active position numbers and equity usage i guess it's something like ~1%?
Yea allocation is controlled from the individual strategy level. An individual strategy may allocate 5% of capital to a position. But that strategy may only have 15% of total portfolio allocation. So that position is only 0.75% of total portfolio capital. If three systems buy the same stock I have a 2.25% allocation. So still pretty low.
Very interesting thanks for sharing. For reducing correlation to market, how about simply maintaining a negative position in spy as percent of portfolio. It looks like you have significant alpha so that will still give you strong returns uncorrelated to market. Realistically it will look like less spy exposure in your non trading accounts.
I’m considering something like that. I may try to be a little more nuanced about it and have some timing so I’m not always short. But could be a decent start towards being more market neutral
makese sense.. btw are you in the US? How do taxes work for trading on TSX and ASX, did you have a lot of headaches last year because of that? Or was it relatively straightforward with no surprises.
Thanks for sharing. Are you planning to manage money for others and build a track record? Please DM me if you have any interest in this.
Anyway. When you are ready to share your track record let me know. I don’t need to know too much about what the code or algo is. I care more about the Sharpe Ratio.
Will keep you in mind! One thing I’ve always wondered: why do you care about sharpe? Not calling you out specifically, im using “you” generically. Like if I could offer a CAGR of say 20% and a drawdown around 20% but a sharpe of 1, wouldn’t that be better than a cagr of 10% drawdown of 10%with sharpe 2 (assuming neither use leverage)?
Sharpe ratio is the gold standard for this industry as long as you are not selling options. Option sellers naturally have high SR. There are hedge funds with returns less than 10% with a 2.5 SR and they have billions waiting to get in. I worked for one of them. Investors like to sleep well knowing that their capital is safe.
I get that. But after fees and taxes, would you just be better off in a 4-5% return money market account or sorts?
I’m coming from the perspective of wanting high growth. So I take some volatility to get that.
Two different objectives, that’s all
If your returns are good you can invest for me and I can pay you 20% of returns. Not much legality if you put the orders in my account through the API. This is called a managed account.
I’ll look into it. My account is pretty active and trades multiple countries so would need to sort that out
I’ve considered it. Currently building a “track record” with my own money as proof of concept. But I have no idea the legalities of managing money.
Thanks for the post, and congrats on the great results.
I have a question regarding higher time-frame strategies (daily/weekly/monthly). I’ve noticed that many traders (including you) successfully run systems on these time frames, but I’m trying to better understand the backtesting and validation process.
When trading on higher time frames, a 10–20 year backtest often results in only 100–300 trades (aka statistically insignificant). How do you evaluate and gain confidence in a strategy with such a limited sample size?
Do you rely less on statistical significance in these cases and more on structural or economic reasoning behind the strategy? I’d be very interested to hear your perspective 🙏
Thank you and good question.
Actually most of my long term system have 800 to 1,000+ trades believe it or not. But not all. I generally test into the 1990s so get a lot more data.
Generally I’ll run a lot of robustness testing to ensure the system is actually measuring the signal and not curve fit to few trades. I’ll force the system to swap out the trades taken in the Backtest with other valid signals generated by the system but skipped due to position number constraints. Like a resample Monte Carlo.
Thanks for the reply. So, do I understand correctly that you aim for at least a 1,000-trade sample size, and if a strategy has fewer than lets say 300-500 trades, you try to increase the sample size? What is the minimum required number of trades for you to consider a strategy statistically reliable?
Yes I try to increase sample size either by letting the system hold more positions or forcing different trades but still viable signals into the system (think testing the system on a similar but different universe or forcing the system to skip a trade and take a similar but different trade).
More trades is always better but I try not to get hung up on something having too few trades. If the system is exploiting some phenomena that I have reason to believe is real (momentum for example) I may be okay with a smaller sample. I’ll just be sure to keep the strategy super simple. For example I have a seasonality system that has ~240 trades in it. But the seasonality measurement is super strong and persistent. I still trade it even tho it has a small sample size. I just make the system super simple with 2 parameters.
Got it! Thanks for such detailed answer, much appreciated, have a nice day/evening!
Absolutely you too!
Sorry if I join this discussion.
Is correct that 300 trades are not so “many”, but you have to consider that in 10-20 years, the market faced every scenario, bearish, bullish, sideways, so if the “system” created 300 successfull trades in 20 years with positive result, I consider it a valid system
Hey 👋🏻 No worries, thanks for your opinion! I agree it depends on context and logic of the strategy (That’s why I initiated this thread😅)
like as Author said, if this is strong seasonal tendency - we might be fine with smaller sample size (because there is some fundamental logic behind the price movement)
However, I would not consider 200-300 trades sample size for EMA 21 crossover on weekly timeframe (on some random market), since this is looks like overfitting + low sample size.
You got it man. That’s exactly right.
Very nice write up! Thank you for sharing.
I do the same with my Mean Reversion strategies. I’m not a fan of Mean Reversion either, multiple strategies spread out the suck.
You run a lot more positions than I do, I use 1 position size formula for all strategies. Max allocation percent vs max ATR volatility. Some strats use stops but fill-to-stop risk is never ask issue with small sizes. On strategies I want more allocation to, I’ll add a x2, or whatever, to the quantity formula in the strategy.
I would like to see your correlations, both returns and dds, for your portfolio. I’m about to go over mine for the year and reduce or cut the highly correlated.
Mean reversion can suck at times. Spreading out the suck is key! I’ll have to look at the correlations this year. I can see visually on the pnl plots there were some days where correlations were 1. Best way to make that smoother is to add things completely different than what I currently have
Thanks for sharing. How do you size your positions in your strategies? Is this fixed percentage, vol target or anything else? Do you have an upper % limit for single position and ticker?
do you trade mostly (leveraged) ETFs or stocks?
It’s a mix of equal weight and vol based. I don’t have an upper limit. If 3 systems say buy xyz ticker I buy it with essentially 3x size. If 3 systems say long that means the signal is strong so I accept the concentration risk. I’m holding 140 other positions anyway. Most of what I do is single stocks. I have some lev ETFs but thats a relatively newer addition.
Thanks. So multiple strategies might enter trades for the same ticker at the same time so there's no limit on that. I guess you do have some limit on each single position for a specific strategy? From your active position numbers and equity usage i guess it's something like ~1%?
Yea allocation is controlled from the individual strategy level. An individual strategy may allocate 5% of capital to a position. But that strategy may only have 15% of total portfolio allocation. So that position is only 0.75% of total portfolio capital. If three systems buy the same stock I have a 2.25% allocation. So still pretty low.
great job - amazing what persistance can do - i love your portfolio aggregation idea and look forward to learning more about it and more from you.
Thank you for sharing
Mike
Appreciate that Mike!
great job - amazing what persistance can do - i love your portfolio aggregation idea and look forward to learning more about it and more from you.
Thank you for sharing
Mike
Great job and thanks for sharing
Very interesting thanks for sharing. For reducing correlation to market, how about simply maintaining a negative position in spy as percent of portfolio. It looks like you have significant alpha so that will still give you strong returns uncorrelated to market. Realistically it will look like less spy exposure in your non trading accounts.
I’m considering something like that. I may try to be a little more nuanced about it and have some timing so I’m not always short. But could be a decent start towards being more market neutral
makese sense.. btw are you in the US? How do taxes work for trading on TSX and ASX, did you have a lot of headaches last year because of that? Or was it relatively straightforward with no surprises.
It was pretty straightforward. The tax forms were auto created by interactive brokers and I just fed those Turbo Tax