Transcription
Every single trader should know these six performance metrics because they form the backbone of how you actually measure successful strategies. They're also the exact same metrics I use to evaluate my own automated portfolio of trading strategies. These metrics have helped me identify what's working, what's not, and where to scale capital across all markets, both in futures, equities, and in crypto.
Now, I've just closed out my November month. So, I'm going to walk you through my actual performance using these six metrics. And you can see my equity curve here overall for the month. These are the metrics that tell you if your trading is sustainable or if you just got lucky. Because here's the brutal truth: You could make 50% in a single month and you'll still be on track to blow up. So, let me show you my November performance over 10 automated strategies and, more importantly, how I actually evaluate it.
But just before that, I want to play you a quick clip from Cliff. He is the a billionaire and co-founder of AQR Capital Management, which runs one of the largest systematic and quant funds.
Literally since our Goldman Sachs days, this is more than 25 years ago, we've generally used half a back test out of sample as a bogey, and life has worked out fairly close to that. Now, we've built more factors, improved our back test, so it's not necessarily half of what it used to be, but that's not been far off. And we would consider that a victory. If you have a decent back test that you don't think is overdamine, that you think is reasonable, that you understand the economic spirit, that you've tested robustly in a whole bunch of places, all of that, still assume half going forward. And half is obviously a rule of thumb. You may sometimes think 2/3, 1/3 makes more sense, but it's good to have a simple thing you don't deviate from.
Now, as you heard in that clip, he says that they've seen around half of their back test results or out of sample results of the back test at G Live. And I think that's a very clear thing to keep in mind with this whole presentation is that most of the time your back test results are not going to fully line up with your live expectations. There is going to be some difference in performance, and that's totally natural. Your back test is a historical test, which is still a great way to validate, but live performance is always a bit more tricky.
So, let's get into the first metric. I think no one was really surprised by this, but it's Sharpe ratio. So, Sharpe ratio primarily just tells you the risk-adjusted returns, and that primarily means how much you're earning per unit of risk taken.
Now, my Sharpe ranges from 0.7 to about 1.2, give or take. And here's how to read it. Because I see a lot of terrible kind of information around Sharpe ratio, and I think most times from people that have maybe done a lot of theory, but haven't actually taken strategies live or tested strategies live. Most of the time, you are not going to get a perfect two, uh, Sharpe ratio. Most strategies are going to be above, hopefully, the 0.75 range for strategies you actually take live. And normally, very good ones are about 1 to 2 Sharpe ratio. If you really have anything above two, you've got a absolutely amazing inefficiency. And most of the time, those aren't going to last very long, but at least for that period of time, it is a great Sharpe ratio for a strategy.
Now, obviously, if you have anything, you know, below zero, then that's obviously negative. Or I would say really even below like 0.5 is not great most of the time. Now, if you have something that's crazy good, so let's say like a Sharpe ratio of three, four, or five, that can sometimes be suspicious if it's overfitting, but it does depend on how the Sharpe ratio is calculated.
Now, some platforms do Sharpe ratios on a daily basis, which doesn't really make much sense, but that's just how they do it. Some do it on a monthly, yearly, etc. Ideally, you should always be looking at your Sharpe ratio on a quarterly or yearly basis. At least when you can, if the platform is, you know, for whatever reason, showing it using, you know, daily trades, etc., then you'll just have to export that data out and then do some custom code around it.
Also, one thing to keep in mind, your Sharpe ratio changes quite frequently. At least mine has. That's why I put a range because, you know, some years or, you know, some quarters I'm 1.2, other quarters I'm 0.7, you know, 0.5. So, it does change, but most of the time, if you're aiming for that, especially if you're doing a historical back test, you want to be looking for the 0.75 plus to ideally like 1.5, if you can get that in more inefficient markets or if you're doing an actual inefficiency and not like a risk premium, I think you can definitely get the two and maybe just above two.
But here's what most people miss around Sharpe ratio: it doesn't actually mean that your strategy is sustainable. You could have a Sharpe ratio of two, for example, and still have ways to blow up that particular strategy. So, you still need to look at other metrics, and you do also need some other kind of tacit knowledge around it.
Now, the next metric is drawdowns. So, I thought here is a great place I'll show you my kind of performance over all the strategies for November. As we can see, my total performance was 6.42%, which, you know, normally I'd be saying that's fine or like pretty good, but my drawdown was 16.69%. That was pro, it was a very tough month for me, I'll be completely frank. My historical average drawdown for my portfolio is normally around 9%. Now, my max drawdown, I think, is 27. Yeah, it's 27, I think, from historical back test. So, 16.69 is definitely in scope, right? It's not unexpected. It's not something crazy happened. It's just a lot of volatility and wasn't a great month to sit through. It was a decent amount of money being down at one point, and that's just how the game is. I still ended up profitable, which I'm very happy with, but it's still a different story, right? If I just told you, "Hey, I made 6%," most people would be like, "That's great. That's an awesome monthly performance." But when you take into account the drawdowns of it, it's no longer as, let's say, flashy, right? It's not as amazing.
Drawdown, in case you don't know, is measured from peak to trough. So, it basically just tells you the worst decline you experienced over that X amount of time. So, that could be over a month, as like for me for my November, or it could be over a year, etc.
Now, the goal isn't necessarily to have a good drawdown because this really depends on your market and your risk tolerance. Drawdown and risk is technically something you can change and adequate to. You can always de-size, which will always reduce your drawdown. So, that is a part of trading that you can manage. You can definitely manage your risk. But to counteract that point, on a market, for example, that is more volatile, you'll normally have a higher drawdown to return ratio most of the time. For example, in crypto, right, it's quite easy to get a really good strategy in crypto, but it's still going to maybe have a 50% drawdown over that period of time, let's say over a 5-year back test, etc. So, it does depend, but ideally, I'd be looking for a 10 to 20%, but this does depend on your risk tolerance, you know, your position sizing, what size of account, all those sort of things. But as I kind of just explained, this matters because a 25% drawdown with 80% annual returns, that's still amazing. I would run that strategy. But, you know, 25% drawdown with 15% returns or 30% returns, that doesn't really seem to make much sense. I wouldn't run that particular strategy.
Now, the third number that many people don't talk about is the average drawdown duration. Most people in their head, when they see a drawdown on paper, they think, "Oh, you went from, you know, down 16% and then you were, you know, plus 6% and it was instant." No, my drawdown lasted 13 days, which isn't that long. But it was still a stressful period of time. That's still a lot of days that you can doubt yourself. It's still a lot of days where you can think, "Hey, is the strategies, you know, no longer working? Did I make a mistake?" All those sort of things. You know, technically over historical average, my drawdowns normally last around 60 days. So, this was quite short, to be honest. It was just a bit more volatile than I really expected. For example, if you, you know, know your drawdowns last 60 days, but you only have 30 days of patience, you're most likely going to cut your strategy too early, or you're going to change it and then mess up the whole thing. So, knowing the average drawdown is very important so that you can actually handle those drawdowns correctly because mental capital is really as important as financial capital. And that's why I'm such a big promoter of like automation and why I try and help traders automate their trading as well.
Now, metric three is CAGR. So, this is primarily just your annual return including compounding. So, instead of just saying, you know, "I made 50% this month," if you can say, "I have an average CAGR of 50%," that is way more impressive than having a single month that does, you know, 50%. And then generally for the benchmark, you know, you're trying to beat normally SPY or whatever your benchmark is for that particular market. So, normally that's around 7 to 9% annually. If you're not beating this often, you are better not running the strategy.
But I will put a huge caveat there that not all strategies are built for the same thing. Some strategies are actually built for just reducing drawdowns in the portfolio. Some strategies are built for portfolios that are, you know, very risk-averse. They want a really small amount of drawdown, but they're still going to have be happy with maybe 6 to 7%, right? They maybe just want less drawdown than SPY, for example. If you could even beat like if you could match SPY performance but have half less the drawdown, you're still an amazing strategy. So, do keep that in mind, 'cause I see a lot of people talk a lot of nonsense around like, "Oh, you should always be beating benchmark." That's not necessarily true.
And also one last thing here to check is like if you're making, you know, 90% of your returns in a back test or even in your live performance from just one year, that's often a red flag. That doesn't mean the strategy doesn't work. It just means it only really works on one environment, or it was only present for that one environment. Maybe that environment comes around again, but that's not a really great thing to just hope for. You'd want to have some metric to see, "Okay, like it's that environment is coming back around, and now I can activate that strategy."
And to give you a bit of, you know, an example, one of the strategies in our crypto momentum group beat Bitcoin only by 90%, which still, to be fair, because of how much Bitcoin has compounded over the past like 5 years, is quite still impressive. But the CAGR is also higher of 57.8% compared to 54. And then it also has a bit of a better Sharpe ratio, and it also has a better drawdown. So, overall, all of those added together, that's a strategy I want to trade, and that's why I'm trading that strategy. Also, most importantly, we'll get to this in a bit, is about correlation. The correlation is very low, which is exactly what I want. So, this is like a it hits it everything out of the park, and that's why it's a strategy I'm running live.
Now, four is win rate. Win rate, I'll get this really quickly, but primarily it's very simple. It's just the percentage of trades that you are being profitable on for that particular strategy.
Now, my November win rate was 39.7% for the whole portfolio. So, you know, quite low. It's around 40%. But that is important to know is that I do have a decent amount of trend following strategies within that. Now, what most traders get wrong here is they just think win rate, you know, without any context. A trend following strategy having a win rate of 20 to 40% is actually amazing. I'd want a percentage around that. If I had a win rate on a trend following strategy that was at 80%, isn't a trend following strategy, I've probably overfitted it, and I'm making some sort of mistake, and that's not going to work long term. I want it to match the characteristics of that particular strategy type. And trend following normally has a lower win rate, but you're going to have more frequent losses, but you're going to have rare massive wins, and those make up the difference. Whereas on mean reversion, you know, I definitely want a win rate that's above 50. I'd want maybe 60, 70, etc. You're betting on small frequent wins and then some very rare large losses. And then if your, you know, if your trend following strategy has an 80% win rate, that's probably overfit, like I mentioned. And then if your mean reversion strategy suddenly drops to 30% win rate, something's broken, maybe that signal is no longer working. That's something I'm kind of scared of.
Now, metric five, and this is probably the most important out of all of these because it creates the the rest of them, is your trade count. What is shown over here is the law of large numbers. And it basically means the more samples you have, the closer you get to the expected value of what's actually happening for that probability. So, you need to have enough trades to know if your results are real or they're just luck.
So, here's the rule of thumb, 'cause everyone always asks me like, "How many trades do I need?" Ideally, you want around 300, give or take. But this does depend on time frame, depends on what sort of signal you're trading, and then also how much time you have, really. For example, like if you're doing a swing strategy and you're only getting, let's say, 20 to 50 trades per year, that's going to be very, very difficult to say, "I'm going to wait four years until I get like enough statistical significance to make any changes." That technically can be a theoretically correct way of doing it. But there's differences here and there based on time frame. But the key point here is do not, you know, base any performance metrics on a strategy with 30 trades or 30 trades when you're live. It's not enough. You're not even going to get close to having anything that's significant, and you could just be looking at a very unlucky or lucky sample.
Now, with 300 trades, you're often seeing more real, kind of like, say patterns within your data. So, you always want to track your rolling sample size. You want to track and understand that one month means nothing. This is the hypocrisy of these monthly updates. I do these monthly updates cuz they're fun for me to record. I like sharing my performance, and also it means that people know that I'm actually trading something, right? But I would say one month doesn't really mean much. If I went, you know, for 5 months in a drawdown, that would definitely mean something. That would be important to see why that's happening. But one month performance like this, I'm not that worried. Obviously, if this one month performance I had a 90% drawdown, that would matter because there's something significant about that that time. But because everything is within expectations, it's just onto the next month, you know, see how it goes. It was definitely not a fun month to go through. But ideally, normally one, two, even 3 months doesn't matter that much. And then once you're getting, you know, above 3 months, give or take, ideally you want to be looking on a quarterly basis or a yearly basis. That's where stuff can really tell you something that's important.
Now, number six is strategy correlation. So, this is my correlation for the past month. This is probably one of the things that I would note from this month. The reason why I had such a terrible drawdown this month is because two of my strategy, oh, actually more than two. It was like two were fully correlated over here. You can see one, and then a couple of them were quite correlated, about the 0.7, and they all went into losing streaks. So, that drawdown was completely, you know, added because of that.
Now, most traders, I would say, completely ignore correlation, or they don't really understand it fully. So, to kind of break it down, correlation tells you how strategies perform together. If you have a correlation of plus one, that means they're in perfect correlation. They're moving in tandem. So, if one goes up, the other one goes up. If it's at zero, then there's no correlation. It could one could go up, one could go sideways. Doesn't really matter. Correlation of minus one means one goes up and the other one goes basically down. So, you're getting negative correlation.
What you want is as close to zero or possible, or even better would be zero negative correlation. Negative correlation is quite hard to get in the market. Most things are correlated nowadays just because we have such a global infrastructure. So, most things do kind of trade together. However, you know, the more you can get on that zero side or even a bit close to the negative, the better. That's where you're going to get all these benefits because you're going to get the benefits of smaller drawdowns. You're going to get the benefits of more consistent returns because you have things uncorrelated to each other, so one can make up for another, etc. You're going to get higher risk-adjusted performance. It's going to be a better Sharpe ratio. And you're going to be able to use leverage safely because you have multiple things playing in tandem, making more consistent returns, meaning you can scale faster.
But let me give you a quick example, 'cause technically this month was an example for me. Let's say you run five strategies, and they all are correlating at 0.85. So, near one. When one loses, they all lose. And then that's going to mean any drawdown you suffer is going to be basically five times as worse because they're all going to go into drawdown. Whereas if you have, you know, five diversified strategies, and one strategy is losing, doesn't really matter. The other ones can make up for it, whether they be negative correlation or zero correlation.
Now, how do you actually diversify? Well, ideally, you want to be diversifying across different asset classes. So, equities, bonds, commodities, crypto, etc. You want to have different strategy types. You want to have trend following, mean reversion, arbitrage, all those sort of things. You want different time frames: intraday, swing, and position trading. You want long and short. So, long only most of the time are going to be correlated with, you know, or during crashes for the most part. Depends on what the signal is. So, ideally, you want some long and short in there.
Bottom line is if you have a strategy that you know has a 60% win rate, 2.5 Sharpe, and a 12% drawdown historically, if they are all correlated at 0.9, one bad week, you're going to get a 30%, 40% kind of drawdown.
So, you might be wondering, you know, what am I doing to change this? This is the caveat here on my particular one. Now, I could totally be wrong. Obviously, I'm still learning myself, but this is only one month, and necessarily I know correlations do spike. You're never going to have the same correlation forever. So, correlations could have just spiked this month. Now, potentially I maybe want to add more strategies to kind of rebalance that of how many I have trend following, for example.
However, because it is only one month performance, and I haven't really ever had before any bad correlations like this, apart from when it was a like a whole market thing. So, like the Trump tariffs, my portfolio did go a bit down, but overall, because I had so many different strategies, they actually evened out quite well. So, you know, I've had more stress events like that, and they've still done decently. So, this was quite a surprise because nothing really happened that crazy to have myself this much correlation. So, we'll see. I'll definitely keep you guys updated if I change it over the time. But that's primarily it.
Now, to give you one thing to remember from this video: no single metric tells you the full story. Please use multiple metrics. Don't ask the question, "What metric is perfect?" There are multiple metrics you need to look at.