Trend Following in Times of Crisis
Trend Following in Times of Crisis – a webinar by Nick Radge for Amibroker Canada User Group. We would like to thank ACUG for the opportunity to present to your members.
7 April 2020
In this webinar we’re going to do is take a look at how I structure things. I do things a little bit differently, I guess, from your regular kind of investor. Rather than invest in a portfolio of stocks, what I like to do is invest in a portfolio of strategies.
I’m essentially been a trend follower and a systematic trader from day one.
I started trading when I was 18. And believe it or not, I started trading using a five and a 10-day moving average on your equivalent of the S&P 500. So that’s all I had, there was nothing else involved in that.
The reason why I trade a lot of different strategies is that I understand that not every strategy will work all of the time. But they all offer long-term out performance when put together.
If you’ve only got one strategy and it goes through a drawdown or a flat period, people can get very frustrated. And as a result, they’ll tend to try and do something different. And I call that the beginner’s cycle. When people flick through different strategies, they give something a go for three months, it’s not really working for them, they try something else, and they try something else again.
So to overcome that, rather than try and have the perfect strategy for the perfect market at that time, what I try and do is have very robust strategies that deal with different market conditions in their own way and, over the longer term, they will have a positive outcome.
So not only do I diversify across strategies, but also styles, time frames, and markets, and we’ll get into that as we go through here.
Firstly a broad overview of my personal investment portfolio. I have a tactical portfolio, which is what we’re going to be talking about tonight. And that makes up the majority of my portfolio and that’s all my active strategies, ASX and US strategic. They’re basically buy and hold positions that we have in a small number of companies. They tend to be longer term in nature. Some of those we’ve held for almost 20 years. For example, Apple, Amazon, Tesla. There’s a couple here in Australia that I have. We have a bond portfolio and a little bit of cash.
But what we’re talking about tonight is the tactical allocation; the active strategies. They are all 100%, systematic, they all run through Amibroker.
My baseline for a strategy is to have better metrics than buy and hold, which, is about 8% or 9% return and a 50% drawdown. So I look for a minimum annual return of 15% and a MAR ratio higher than .8. So the MAR ratio is where we divide the annual return by the drawdown, so I want that to be about .8 or higher.
I’ll run through my strategies and give you an idea where these ideas have come from, what they do, etc.
The ASX Growth Portfolio has been my mainstay. I’ve been trading it for over 20 years, it’s the same strategy that I created back in the ’90s. It’s hardly had any changes to it in that period of time. It’s based loosely on a Bollinger Band Breakout trend following strategy. I covered that strategy on my book Unholy Grails. The Growth Portfolio is loosely aligned with that.
I’m a big believer in the long-term application of a simple strategy is what creates success. The Growth Portfolio is the full allocation of our retirement account. However what I’ve done now is divide that into two different strategies and we’ll talk about the reason why a little bit later on. But the second part of my retirement account, it’s going to be looked after with the Weekend Trend Trader. It’s a weekly breakout strategy. Again, it’s pretty simple. I trade this one a little bit more aggressively. But the whole idea is to divide that portfolio up in two to get some diversification. So the Growth Portfolio and Weekend Trend Trader are trend following strategies.
Then I trade two momentum strategies. So they are relative momentum; the US Momentum and the Trade Long Term Premium Portfolio. They trade monthly and are a simple rotation strategy to keep you invested in the strongest stocks at any particular point of time.
The US high-frequency, I started trading this back in 2010 after several years of research during 2008 GFC when we were sitting on the sidelines. This is a short-term multi-day swing trading system, holds for three or four days on average. Basically a mean reversion, long only mean reversion strategy.
Then, the day trade strategy is actually made up of three separate systems. This is fully automated, trades the US market through the night, and that one is the strategy at the moment that’s doing particularly well.
And the market-neutral strategy, this is under-researched. It’s a live test account at the moment. It seems to be holding up okay for the moment.
If we look at the style allocation that I currently have, trend following makes up the most of it. It’s about 70%, mean reversion is the other 30%. Now, I’ve got market-neutral in there, but it’s actually a trend following market-neutral strategy. It sounds a little bit odd, but it’s something a little bit different that I’ve never really seen anyone else do. So I may be barking completely up the wrong tree with it. I’m about to find out with the real-time trading to let me know what I’m doing wrong, but I’m excited about it.
I guess for a lot of people, this looks overly complex. It’s not, it doesn’t take long to operate on a day-to-day basis. In fact, some of these portfolios operate monthly, so there’s really nothing that has to be done. There’s the style breakdowns, they’re absolute trend, relative momentum. The intervals of which I trade them, so the Growth Portfolio is a daily. The Weekend Trend Trader’s, weekly. US Momentum is monthly. Premium Portfolio there is another momentum strategy that trades both monthly and weekly. High-frequency’s daily. Intraday there for the day trade, obviously.
Current status of those: obviously the trend strategies are in cash and year to date performance with some of those strategies at the moment is ordinary. The US high-frequency that’s our multi-day mean reversion strategy. We got smacked in February 2020 and then pulled the leverage off in March. So that’s where we’re at with that. Obviously, the fall in February was pretty sharp, and we held on to a couple of those positions. It’s one of those things that happens. I’ll be talking about how to deal with drawdowns.
You can see the day trade strategy there and the market neutral strategy are both ahead year to date. So that’s where we stand with those.
Let’s discuss managing tail risk with trend and momentum strategies. I approach it in a variety of different ways. The first line of defense for me is a pretty stock standard regime filter. A regime filter is basically giving us the direction of the broader market. In this particular case, we’ve got the S&P 500. I use the same regime filter for every trend and momentum strategy that I trade. In its simplest format, signals are only generated when the regime filter is bullish. So let’s assume our regime filter is a 200 day moving average of the S&P 500. If the S&P 500 is above the 200 day moving average, then we suggest the market is bullish.
Our momentum strategies will exit all positions immediately at the end of the month or at the end of the time frame when the regime filter turns bearish. So again, if we use this example, if the S&P falls below the 200 day moving average at the end of the month, I will then exit positions. The absolute trend following strategies do something a little bit different though. What we do here is when the regime filter turns bearish, two things happen. One, there’s no new buy signals. Just remember the old adage that a rising tide will lift all boats and a falling tide will tend to drop all boats. So even in a sustained bear market, you may get one or two stocks that buck the trend and go up. We’re seeing that at the moment. There’s a few stocks including Zoom that’s going up, but in a broad market sense over the longer term, most stocks will go down.
So in that particular case, I don’t want to be involved. So the strategy will stop generating new signals if the regime filter is bearish, that’s the first thing. The second step and this only applies to my absolute trend following strategies. The second step is that I tighten the stops up, they ratchet higher. As a basic example, let’s say we’ve got a trailing stop that sits 20% behind the current share price. When the regime filter switches on or if we determine it’s a bear market, that trailing stop ratchets up to, say, 10%. What we’re doing here is we’re basically keeping the door open, if you like, where rather than shutting the door and going straight to cash, we’re going to keep the door open a little bit. The reason being is my research shows that many of the leaders that we’re holding tend to pause during declines rather than drop, and if they pause during a decline, you really don’t want to exit that position, ’cause they’ll consolidate and then when the market turns off again, they’ll be the first to jump out again.
So what we do is we leave a little bit of room for movement there, but we don’t let the full stop go, we don’t let it go to 20%. And what that does, the regime filter where these ratchet stops tends to greatly enhance the risk adjusted return. And I’ve not found one trend strategy anywhere that can’t be improved, at least from a risk adjusted return basis, on doing this kind of exercise with a regime filter.
If we have a look here, I’ve basically taken a momentum strategy with the regime filter on, that’s the blue line. You can see there was a drawdown of around 20% during 2008 GFC, and then it went to cash. If we take the regime filter off, we can see that we still get a good return over the longer term, but we go through that deep drawdown of 46.8% during the GFC.
Now, my experience suggests that the average retail trader can handle a drawdown of around 20%. So a lot of the strategies I build for my clients, I build with a drawdown of around 20%. There’s no point building a strategy with a return of 40% and a drawdown of 40% if you can’t handle a drawdown of 40%. Everyone wants to make money, but the key is you have to get through the bad times in order to make the money. If you can’t get through the bad times, you shouldn’t be trading in in the first place.
So, at least with systematic trading, we know what the bad times are going to look like and we can adjust the strategy so it fits within our risk profile.
During times like this [COVID19], and more so when your trend systems are sidelined, it’s an ideal time to research new ideas and think deeper about what you’re doing and how you can do things better. And as I said back in 2008, up until then, I was only trading trend following strategies, but I took the time while we were in cash for most of 2008 and early 2009, to research mean reversion style systems, swing systems, shorter term in nature. And one of my goals was to identify a strategy that could operate profitably in bear markets, and we have successfully done that.
Now, these are long only strategies. A lot of people ask, “Why don’t you trade short?” So back in 2008, in Australia, they banned short selling and they did so for quite some time. Now, the problem is that if you’ve got a strategy that you rely on that trades short and you can’t use it, well, it’s a waste of time. I note this time around again, we’re seeing some conversation out there asking for short selling to be banned again in Australia. There was some academic white papers done after the 2008 situation that said, it made no difference to the bottom line, and I agree with that. But at the end of the day, what it made me do is say, “Right, how can I profit trading long only when the market is trending down?” We’ve managed to do that reasonably well, I think.
So whilst regime filters are the first line of defense for trend following strategies, there’s also three other areas that I’ve been working through. Some of these have already been implemented and they’re running and others are still a work in progress.
The first thing is the diversification of strategies. I’m trying to diversify style, so I’ve got absolute trend of relative momentum, I’ve got mean reversion, and I’ve got day trading, so lots of different styles in there.
Second of all, global allocation. At the moment, I trade the ASX and I trade the US. Ideally, I’d like to trade some other markets, but as Dave mentioned earlier on, Norgate Data only has Australia and the US and I’m not prepared to go to another data supply that doesn’t have survivorship bias or free survivorship bias data as yet.
Thirdly, market allocation. To give you an idea, my US Momentum strategy holds a diversified portfolio of stocks in the Russell 1000 but my Trade Long Term Premium Portfolio trades a concentrated portfolio in the NASDAQ 100, so I’m trading two completely different sections of the market in two very different ways. One’s concentrated, one’s diversified, and as a result, even though they’re serially correlated, you’re still getting a different correlation of returns over the longer term.
Then next type of diversification that we can see here is the duration of the strategies. So the Growth Portfolio, for example, has an average hold period of around eight months. Then we go all the way down to the day trade strategy, which is in and out in one day so you have no overnight exposure.
Another diversification that’s going on here that may not be evident is the trade frequency. We’ve got slow strategies such as the trend following strategies and we’ve got fast strategies, such as day trading, swing trading systems.
We all know that profitability is made up of how much you win when you win and how often you win, but trade frequency is also a big factor. These short-term strategies do a lot of trading very quickly. They all operate on a small edge but we do a lot of trading. We can do up to 200 trades in a day. And as I said, it’s all automated. I don’t have to sit there through the night and manage things.
Then the last bit of diversification is the return path. You can see those different strategies had different return paths depending on what’s going on. All the longer term trend following strategies are now sitting in cash or this is the GFC, but you can see a lot of them went to cash early there and I’ll wait and that’s difficult for some people and that’s another reason why having other strategies there to stop you fidgeting and doing other stuff, but we even knock that down.
Market allocation: the day trade strategy, for example, that’s actually two different strategies that are trading two different parts of the market. So one system trades the Russell 1000 and it also separately trades the Russell 2000 constituents as well. The same system, the exact same rules, trading two different parts of the market. Now the reason why is because of trade frequency. If I can trade that strategy in the Russell 1000 and do 1000 trades a year, well, I might as well trade it in the Russell 2000 and get another thousand trades out of it each year, which is exactly what it does.
As I said, the exact same strategy, exact same rules, exact same parameters, everything is exactly the same, just trading on two completely different parts of the market. You can see you can get some diversification there and that’s the whole idea.
Let me give you a basic example. Let’s say our regime filter switched off at the end of February, which is what mine did, and I exited all positions on the first day of March. Let’s say you had a different regime filter that didn’t switch off at the end of February and made you hold through what we’ve seen in March. Now, I can tell you the difference is a drawdown in situation one of about 10% and a drawdown of something like 40% in situation two.
Now, if we’re using the same kind of strategies, the timing luck of that regime filter is what’s done the damage. So to give you another example, some of the pension funds are actually refusing to do their quarterly rebalance due to market dislocations. They’re making the assumption that they will steady the ship and everything will be good next quarter, but what if it’s not?
So my point is that if you’ve got a single regime filter trigger, let’s say, it’s a 200 day moving average, because it doesn’t have a lot of frequency and triggering, you have an increased risk of running into this situation that we’ve just basically been through. But if you have two different filters, so for example, if we take that strategy and we use the monthly one, and then we take the exact same strategy and rotate it on a weekly basis, basically, what we’ve done is we have lessened or diluted the timing luck of that regime filter, because there’s more examples or there’s actually more rotations. We’re rotating every week instead of once every month, we can remove a significant part of that or dilute a significant part of that timing risk, if you like.
The same would go for my ASX trend following strategies in my retirement account. If I have my whole retirement account in one strategy, then I’m at risk of that strategy having a timing problem. If I then add a second strategy and put 50% of my retirement funds into strategy number one and 50% into strategy number two, then basically, I’m removing or diluting the timing luck of one of those strategies.
Here on the screen, you can see the monthly and the weekly and you can see straight off hand that the monthly actually performed significantly better than the weekly, this is a log scale chart. So the difference between the two is reasonably big, but if we then put the two together and allocate 50% to the monthly and 50% to the weekly, what basically happens here is our annual return goes from about 24%, down to about 22.5% and our, which is not too bad, that’s the payoff that you’re going to get for it, and you drawdown goes from about 19.5% down to 19. So you get a benefit in terms of removing and diluting the timing luck, and that’s the most important thing.
There’s other ways you can do it, instead of rotating monthly, you can take the same system and rotate it weekly. What I’ve actually done is I’ve taken the same system, and I’m rotating half each month. The same system, I’m applying weekly. For example, let’s say, I was using a 200 day on the monthly, I’m going to use a 100 day on the weekly. So not only am I changing the timing of the rotations, but I’ve also diversified the length of the regime filter as well. So the yellow line there gives you an idea of what the equity curve would look like, and that removes those kind of risks.
I’ve been thinking of a hedging strategy, a hedging tool. My experience with hedging is that it can be expensive and it can be a frustrating exercise. If the hedge signal is too tight, you can get whipsawed badly.
Now, I’m sure most of you have heard of Tom Basso, a world renowned trend follower. He runs a hedging mechanism and he’s made it all public, but in my view, I would suggest, and I haven’t seen the data, but from my experience, I’d suggest that he would get a lot of whipsaw, and over time, that whipsaw does cost you a lot of money. So that’s one of the things that we need to overcome. We wanna protect ourselves from the big move, but we don’t want the signal to be too tight, so we’re in and out, in and out.
But I’ve seen some hedging methods actually morph into their own trading system, and that defeats the purpose. It’s important to understand that a hedge is not designed to make money. A lot of people get upset, “I’m not making any money from a hedge.” It’s not designed to, it’s designed to neutralize downside risks to some degree.
So my idea is that if I’ve got a monthly rotation strategy, and we get a big sell off in the first week of the month, you’re going to sit there for the next three weeks and just let it roll. That’s how you follow your system.
But if we could then say, “Right, I’m going to hedge. “I’ll hedge those three weeks out “and if we’re below the regime filter “at the end of the month, I exit positions, “unfold the hedge, “and we’re protected the downside to some degree.” So what I’ve been researching is that reactive hedging is too slow. The damage is done too quickly to react to it.
So basic example, let’s say, Tom Basso, I think he uses Keltner bands. Essentially, he’s looking for downside breakout through a Keltner band, and that will put his hedge on, but that’s reactive and my experience and research shows that’s actually too slow. So what I’ve been working on is a predictive way to hedge and I want it to be systematic.
What I’ve done here, and again, I’ve not seen it anywhere else and it is a work in progress is I’ve combined several breadth indicators and a trigger and it is specifically designed to be reactive. So the red bars here are warning signs. It’s saying that the market’s getting overheated and be prepared for something drastic to occur. And then the red arrows point to the actual trigger saying, “Right, hedge on.” Now you can see for the most part, it works reasonably okay and reasonably quickly. The big sell off this year, it came a little bit early.
But you can see it’s still a work in progress there, October 2018, no signal whatsoever. So that’s the one I got slapped with bad luck back then, and that’s why it’s a work in progress. So this is something that I’ll continue to work with, I want it to be completely systematic, I don’t want any discretion, I don’t want any emotion to come into it.
I think, apart from trading a complete diversified portfolio of futures, what I’m doing is reasonably good. Portfolio keeps moving in the right direction most of the time. It’s not 100% perfect, but next thousand trades, that’s my basic view of the world which will keep on rolling and the good times will come.
Trading during a downturn
So the whole idea here with the trend strategies, we’re now in cash. The day trade strategies do particularly well during these bearish environments, and they can knock out some double digit returns and that kind of stuff, which I’d be very, very happy with. What’s working now? Day trade systems, absolutely without a shadow of a doubt. I’ve had 10 profitable days in a row with my S&P, Russell 1000 day trade system. It’s up just over 5.5% year to date.
Admittedly, I took two weeks off, I was traveling in Japan, and it did well over that period when I didn’t trade it. So there’s no overnight exposure with these day trade systems, you’re in and out, and they tend to benefit from overreaction on the open. If people see night futures, they’re down limit down, and I tend to come in on the open market and open and sell and that’s where these mean reversion systems do quite well during the day.
I probably would like to research a few more and get a few more, it’s quite comfortable to go to bed. It’s a very different feeling when you go to bed here in Australia and futures in the US are limit down and you’re thinking from a trend following perspective, “Oh, that will hurt,” but then your day trade system thing, “Oh, it won’t be too bad. “I’m actually looking forward to that,” ’cause that’s when you make money. So you’ve got this backwards and forwards going through your head.
So day trade systems have a think about that. They don’t have to be too complex. The way I do them is I’m looking for stocks that are well oversold. I look mainly at volatility setups rather than standard RSI or Bollinger Band or those kinds of things. And I do things a little bit differently as well. I think, if you use common indicators that are well-publicized, well-published, I think they’re more prone to breaking down in the future. If everyone’s doing the same kind of thing, and I don’t wanna say RSI too is one of those things, but that’s kind of my drift.
To give you an example, my ASX trend following strategy which uses Bollinger Bands. They’re not normal Bollinger Bands. And as a result, you get a very different signal. They’re based on the volatility like Bollinger Bands are, but they’re not out of your box Bollinger Bands. And the only reason I do that is because it gives me a different signal to what it would give anyone else that’s pulling it out of the box. So same kind of thing with RSI. Think differently.
And if you can find something that’s reasonably robust, then you’re onto something there. Another question on the forum was dealing with drawdowns.
Dealing with drawdowns
Drawdowns: you’ve got to suck it up. You can’t not have drawdowns, it’s part of the game. What you want to do is build a strategy that fits your risk appetite completely. The more you attempt to avoid drawdowns, the worse off you will be. Do things like I’m doing, trade multiple strategies, allocate less to a strategy if the drawdown is a little too steep, but don’t change the strategy. Don’t optimise it to fit the drawdown.
Here’s a good story. I was called into a meeting with a friend of mine and there were three guys and they had come up with this algorithm, the scalping algorithm to trade the market using artificial intelligence. Now, these three guys, obviously very smart, never traded before in their life. They had this system, which they kept saying had a 95% hit rate. And I said, “Well, that’s fantastic, “but what’s your win-loss, “what’s your returns and all that?” And they just kept harping on about this 95% win rate. So it turned out their average winning trade was about $27. Okay, remember, they’re scalping, they’re in and out 2,000 times a day. Average winning trade was about $25 – $27. I said, “So what’s your worst losing position?” He said, “Currently we have a position on “that’s down by about $6,000.” So it’s like your whole win-loss ratio is totally blowing out the door.
Now, the problem was, he actually said to me, “Well, we’ll re-optimise the system “so that trade will actually come out “and won’t be there in the future.” And I looked at him, I thought, “You’re an idiot.” I said, “Don’t you understand that there’s just going to be another one of those that comes along down the track?” And he didn’t seem to understand that. So I’ve lost contact with him.
I don’t know how he’s done over this period of time, but he’s totally missing the point when it comes to expectancy of drawdowns and all those kind of things.
My advice is build a system that suits your own risk profile, and then just deal with it. Okay, don’t try and avoid it, you can’t avoid it. So what’s my personal threshold for drawdowns? Look, again, I build systems around that 20% mark. That’s kind of what I’m comfortable with, but I can weather more. I must admit I’m getting a little bit older, so a little bit more conservative probably than what I was. I’ve had systems go into a 30% drawdown mainly from a problem rather than a grind lower, if you like from the system itself. One of the things I am doing with my own portfolio is just moving more from the trend following across to the shorter term stuff. So currently, as I said, I’ve got a 70/30 setup there. I’ll probably want to move that to a 50/50 setup.
When would you remove a hedge?
Once a hedge has been applied, when would it be removed? So it would be removed when the positions in that portfolio are rotated out. So if you’ve got a monthly portfolio, that is only going to take action at the end of each month, then at the end of each month, so long as those positions had to be exited, you would then remove the hedge. That’s the way you would do that. So the whole idea is to protect you during that rotation period where you are sitting on the sidelines.
For your day trading systems, could I ask what time frame are you working with daily, hourly bands?
Are you using a broker coupled with another program? So we are using daily bands. The way the system works is quite simple. We’re looking for a pullback and we will have what’s called a stretch. A stretch would be the daily volatility over the last three, four, or five days. You would take some type of percentage of that, so say take the five-day ATR divided by two, and that’s called your stretch. You would subtract that number from today’s low, and that would be your buy point tomorrow. So what we’re doing with the stretch, it’s like a rubber band. We’re looking for an oversold point, and then we’re looking to stretch it out at little bit further the next day. It makes a big difference to mean reversion style strategies by doing that. So we don’t just enter mark on open tomorrow, do anything like that, we’re waiting for prices to fall even further the next day. So I use daily bars, and one of the things we always do especially with the short-term systems, we check the real-time fields with the back test every single day to make sure that we’re matching it up.
Again, a story from the trenches, I guess, about 20 years or so ago, a guy here in Australia who will remain nameless, he sold a high-end Hang Sang day trading system. And it was an intraday system, and it actually required intraday data to make it work effectively. Now, he wasn’t using intraday data ’cause back then intraday data for the Hang Sang wasn’t available He was using intraday data and he was trading it and testing it in TradeStation, and he went into max drawdown within about two weeks of launching the system. He was charging, I think, $4,000 to sell the system and he sold 200, 300, 400 systems very, very quickly. He made a fundamental flaw in his logic and it was supposed to be picked up by what’s called bouncing ticks in TradeStation, but you can’t test systems like that on the end of day data. So the question is you have to make sure that you’re doing your testing correctly.
It’s the old garbage in, garbage out. If you put garbage in, well, you’re going to expect garbage back out again, so all our stuff is robust in that nature and we also check every single day that our fills are exactly the same as what the back test report is showing.
Now just going on with Marcel’s question there is yes, we use AmiBroker to generate the orders and we have a custom API, which I had built and a custom API is what manages the trades for me during the night. I don’t use AmiBroker, I don’t connect AmiBroker to my AWS platform. I have an API that’s built specifically for my needs and wants and it’s seriously robust and works very, very well. It can hold up to nine systems at once. So that’s the technology I use for that. And that runs on a VPS. It’s a state side VPS. I think it runs in New Jersey. I don’t need the latency, but this is a trader-specific VPS that deals only with traders. As I said, I had a technology problem back in 2013 and that was actually ’cause my VPS decided to reboot halfway through the trading session. So that doesn’t happen anymore with this one.
How do you decide when to use leverage and when to take your foot off the gas and with regard to my sales question, are those limit orders?
Yes, day trade systems, we use limit orders. They get placed into the market, get filled during the day. In terms of the leverage, the day trade systems have incredibly small drawdowns, so they open themselves up to being levered higher. So for example, if we weren’t using leverage, these systems got about 1% or 12% return with a drawdown of about 4% or 5%, so it makes sense to lever them up to some degree and we can do that more to get the high-frequency or the higher frequency trade going through.
The day trade systems, we do operate at your highest day trade margining level which is one, 25% LVR. The swing systems, we operate at 50% LVR, which is obviously the maximum overnight. So recently, I switched the leverage off of the swing system completely. We’re just trading that on a cash basis. That was a discretionary override and that came simply because I was struggling to sleep with that exposure on.
So I know people don’t like to hear it, but at the end of the day, you’ve got to make a common sense call and common sense. Back in February, when the possibility of a 1987 crash was very, very real, then that was just a common sense thing to do. Just turn it down.
Do you use any machine learning?
No, absolutely not. I’m not a believer in that in any way, shape, or form. I’m happy to be convinced otherwise, but I’ve never come across anyone that’s really over the longer term done anything decent with that kind of stuff. So happy to be proven wrong. But no, I don’t do any machine learning whatsoever.
How do you gauge and decide when a live system is not working anymore and when to turn it off and bring it back to development or scrap it?
Well, first of all, you’ve got to build them robustly to start with. We do a lot of stress testing, we try and break the systems. Rather than make a system work, we try and break it. So to give you an idea of some of the stress testing we do with strategies. Let’s say, you’ve got a momentum strategy or rotational strategy that trades once a month, what we’ll do in that particular instance is, and let’s say it takes the closing prices of all the stocks, what we’ll do is we’ll randomise the closing prices. As they say, the future is not exactly the same as the past.
So how, a little bit to kind of replicate what the future might look like.
Well, one of the ways you can do that is you can use data variance. So we’ll randomise the close price of all the stocks by 1%, 2%, 3%, or 4% or 5%, and then we’ll run the strategy again. If the strategy still holds up, well, you’re probably onto something. The other thing is we don’t do any optimisation. We trade the same rules on every single stock, so if you’ve got a strategy that trades the Russell 1000 constituents and all delisted stocks of that Russell 1000, that’s a big universe of stocks, and if you use the same parameters across the whole lot and it comes back profitable, well, you’re probably onto something that’s reasonably good.
I don’t trade any single market systems whatsoever. I don’t trade an S&P ATF system on its own. I don’t trade a Gold ETF system on its own. They’re all portfolio-based, and everything must use the same parameter setting. Other things you can do, for example, I mentioned data variance, so you can vary the signal as well as an example. Let’s say you’ve got a regime filter in your strategy, and it’s going to look at trading decisions on a week-to-week basis. What you can do is add variance to that of 1%, 2%, 3%, 4% or 5%. So that regime filter can be adjusted by up to 5% randomly every week, which in turn will randomise the signals. And again, you would run that, and if it comes up profitable, well, you’re onto something.
We run optimisations to test the robustness, not to pick the the best method. So again, if we use an example of a regime filter, let’s say it’s 200 days, what I will do is I’ll optimise every length from 100 through to 300 days, and I wanna make sure that there’s no mountains in that or anything like that, there’s no roller coaster. I wanna make sure that they’re all reasonably profitable. So it doesn’t matter if the market changes slightly in the future, that somewhere between 100 and 300 days, for a regime filter is going to be profitable.
So that’s our process when building strategies. There’s no optimisation, there’s no single market systems. We try and break the system, try and disprove it, do all these different kinds of exercises to do that.
But coming back, how do you know when your system is broken?
Well, I can’t say if I’ve had any systems break per se, I’ve stopped trading some systems for a variety of reasons. So for example, I used to trade a variant of the Turtle Strategy a long time ago in the Australian stock market. I had to stop trading it because my slippage was costing me about 11% per annum. So I scrapped that simply because the liquidity in the Australian market was such that my slippage was way too high and I couldn’t track that in the back testing. Another one, for example, my day trade systems that I trade in the US, yeah, they work perfectly well in the Australian market, but again, liquidity there, especially when I’m placing limit orders into the market doesn’t allow me to get my full fills. So as a result, your back testing would say you’ve got to fulfill, it’s the lower the day and you got, 50% of your fill and nothing else. And of course, they’re always the best trades. So again, it was not a matter of the system being broken. It was a matter of the count size too big for the liquidity in that market and I just pulled out of that.
My main trend following strategy that I trade with my retirement account, it’s the exact same strategy since I built it back in the ’90s. It has hardly changed at all in that period of time. Look, it doesn’t hit home runs every year or anything like that, but it’s not designed to. It’s designed to quietly just chip away, keep you on the right side of the market, keep you out of sustained bear markets, beat buy and hold, and allow you to make easy decisions without having to read company reports and all that kind of stuff.
The longer term momentum and trend strategies, you have to buy on strength or buy short-term weakness within the longer term uptrend. All ours are based on strength only. We don’t have any trend or momentum strategies that look to buy on any weakness. I haven’t really done too much testing on that. I know there are a few of those type of strategies around, but no, we don’t use that all on strength only for us.
Do you have a systemic way of coming up with capital allocation for each system?
No, again, as much as I’d like to be systematic, it’s more of a gut feel. As I’ve said, over the years over the last four years, I’ve moved more and more towards shorter term. I’m not sure if that’s just an age thing or what. I wanna get to a 50/50 allocation of trend following and short-term. So no, I don’t have any systematic way of doing that. I get a feel of a system that I’m really, really comfortable with. And I’ll probably allocate more to that. But again, it’s about trade frequency, smoothness of the equity curve, what it’s designed to do, the short-term systems are designed to trade in the bear markets. So during this time, I want a higher allocation to those. I don’t move money between systems for different parts of the cycle.
So now, for example, I’m not going to move money from my momentum systems into those mean reversion systems. I don’t do that. I allocate, stick with it, and adjust the allocation as I see fit over time.
Do you typically find better trading opportunities with the Australian market compared to the American market?
Not necessarily, the Australian market tends to trend probably a little better than the bigger cap stocks in America. We’re not as big as those so they do tend to trade better trending because they’re not as institutionalised but then again, the US market is also much better to trade short-term, much more volatility, much more opportunity, if you like. What I tend to find absolute trend following systems tend to work better in Australia than absolute trend following systems in the US, momentum systems work nicely in the US. Mean reversion systems work both but Australia being a little bit more liquid and especially more expensive. It becomes challenging to trade the Australian market, put it that way.
Can you elaborate a bit on the expectancy curve from your book “Unholy Grails” and do you use any external Monte Carlo tools?
Oh, that’s a good question, yes, I do. So the expectancy curve is something I created many, many years ago. I noticed it gets replicated quite a bit by people now. It basically suggests that your win-loss ratio is directly related to your win rate, okay? It’s a curve that it tends to follow. The higher your win rate, the lower your win-loss ratio will become and vice versa. The lower your win rate, the higher your win-loss rate will be. So for example, if you have a trend following strategy that has your standard 45% win rate, then your win-loss ratio will be up two or three to one.
Whereas if you have a mean reversion system, you probably will have a higher win rate. And your win-loss ratio will naturally be lower. So for example, my current stats, I think for this year, I don’t have the trade-by-trade stats, I’ve got the day-by-day stats, day-by-day, I think I’ve had 68% winning days, and the win-loss ratio is about one to one. That’s probably what it will be over the longer term. One of the reasons why this happens is because trend following strategies tend to use what’s called a risk management edge, and what that means is you’re deliberately cutting losses and deliberately letting the profits run. So you are deliberately forgetting about your win ratio and you are concentrating only on that win-loss level. So that’s why our trend following strategies operate at low win rates.
Whereas mean reversion systems, you tend to need to be more, there’s not enough time to have a high win-loss ratio, so you have to focus on a high win rate. So it becomes an entry edge rather than a money management edge. Therefore, you tend to get higher win rates. The problem with a lot of short-term systems is they tend to get optimised very, very quickly, so people want these high winning systems. There’s some terrible, terrible examples out there of pure data mining and that kind of stuff. Larry Williams’ book, “Long-Term Secrets “to Short-Term Trading” is just your classic example of data mining and systems that are bound to fail because they’re simply not robust to start with.
Going on to your second part of the question there. Any external tools for Monte Carlo analysis?
One thing that we do do is what’s called trade skipping Monte Carlo. The whole idea there is that if you don’t use any ranking in your signal generation, for example, we can put this trade skipping Monte Carlo simulation in and what it basically does is we add a random C to the trade selection process. So let’s say on any given day, you’ve got 20 signals, but you can only take 10. The question is, which 10 do you take? Now the best way to overcome that, obviously, is to use the custom backtester, rank your signals and drive ’em through the custom backtester. But another way you can test the robustness of your strategy is say, “Right, I’ll randomly take any of these signals,” and you do that each day, and then you can then plot, 50, 60, 100, 1,000 equity curves against each other on these random signals. Again, if the cluster of those returns are close, then you’ve probably got a system that’s reasonably robust, so that’s a way we can do that. That’s driven out of me broker itself, we have code for that. It’s pretty simple to do and it just gives you a very different look, rather than the bootstrapping that’s in the AmiBroker itself.
I’m not a big believer in the bootstrapping Monte Carlo especially when we’re trading serial correlated equities like we’re doing. I don’t think That really makes a great deal of sense but then again, I’m not a mathematician or statistician, but the trade skipping stuff does work quite well in that respect. I mentioned 15%, see our threshold for a single system. What on average, does your multi-strategy approach do? On average, well, in the last probably three or four years, the trend strategies have been tracking really below their long-term averages. I think and there was two reasons for that. One, this downturn, obviously, and second of all, the downturn in Q4 2018. So the combination of those two hits has pulled the long-term average of those trends strategies below par.
Over the long-term, I think I’ve been tracking it around 15% or 16% per annum. Again, I’m not trying to hit it out of the field. I’m just trying to just chip it away, let compounding do its thing over the long-term, and just keep rolling the systems along, that’s the goal. I’m not one of these guys that can go out and knock out a return of 140% or go into any of those competitions. It doesn’t interest me, it’s just slowly steady, stay in the game for the long-term. I learned very early, I blew up in 1987. My father had to bail me out, so it’s slow steady wins the race, and that’s kind of what we’re running out.
When you say you don’t optimise, do you mean you don’t optimise at all? Or that you do optimise, but you choose a plateau value rather than the best value?
I believe, rightly or wrongly, that there is usually one parameter in your set of parameters that will have the biggest influence on that strategy. You should start with a linear parameter setting to start with to test the robustness of that, you would optimise and choose a plateau value, absolutely. But you would only then go and choose the best value of that one parameter that makes the big difference. So for example, with, say, a rotational strategy, that parameter is usually, the regime filter length, as an example. So that’s really the only one that you would continue to optimise on a year-to-year basis to really make it fit current market environment. When I say that, you only ever want to do that once a year, and you want to do it on the last five years worth of data. You don’t want to do it five times a year in the last three months of data, that’s just overfitting.
On a mean reversion system, usually, it’s the length of your stretch, that is the one that you would continually look at and optimise to fit to the current market environment. And again, you wouldn’t do that any more than twice a week a year and you would only do that on the last three or four years worth of data. You wouldn’t do it on the last three months of data or anything like that. We look at the parameters from the perspective of are they robust, okay, rather than, hey, what’s the best parameter? And we certainly don’t go and change them all the time.
Very, very rarely, as I said, my long-term trend following system that I’ve been running since the ’90s. The only parameter we’ve ever changed is the length of that index filter, and over time, that’s gradually got a little bit wider, not much, but a little bit wider.
Can you please clarify the allocation you use over time?
Do you rebalance to get back to the target allocation on a regular basis when one system outperforms and therefore ends up overweight in the portfolio?
I’ve kind of just done that. My retirement account, for example. I just found that the exposure to that single system was just simply too large a risk and I’ve split the portfolio into two and introduced that secondary system in there. So I guess there comes a point in time where I think, “Well, there’s just too much going on there,” especially from our retirement account. That’s the biggest portfolio I have. So it probably has come, once it’s come to light, the drawdown in that portfolio is only about 13% or 15% at the moment, but it could have been somewhat more disastrous if the timing lock on that was a little bit out. So that’s kind of made me wake up and say, “All right, you need to diversify that a little bit more.” But again, it’s not a systematic way to allocate. It is discretionary, and I don’t know of a systematic way that would really make a difference in doing that.
With the expectancy curve, I came to same conclusion after years of trying to find the best of both worlds. It took me forever to realise that it was a trade-off. Yep, always a trade-off, there is always a trade-off, okay? Absolutely, there’s always a trade-off, and it’s, if you adjust one part of your system, you have to expect it will impact somewhere else. So if you’re going for a higher annual return, you can probably expect some kind of a higher drawdown, if you increase trade frequency. Well, yeah, you’ll get a higher return, but you could also get a higher drawdown. So just be aware of what part of the system you’re trading, what impact that will have on the bottom line.
How do you deal with currency risk for your clients and yourself in trading different market currencies?
What I’d do in that particular situation, obviously, I’m Australian, I trade in the US market when we’re holding US dollar assets. The risk for me is if the Aussie dollar goes up, at the moment, the Aussie dollars of going down, has been for quite some time. So I’ve been reasonably happy with having US dollar assets when that’s been occurring.
However, what I will do is hedge the Aussie dollar and I’ll use futures contracts to do that, so we just basically dollar hedge, simple as that, nothing special about it. I don’t do any, the hedge tool that I use for that is a breakout of 1000 day high of the Aussie dollar, and then I will just hold that until either the positions in the US have been terminated, or that trend in the Aussie dollar turns back down again. So again, keep it pretty simple and that’s the way I do it.
Have you investigated various approaches defining market regime?
I have, in a simplistic way. I’ve investigated using ADX and RSI, those kinds of things. But I just find simple works best, I don’t agree, and I’ve seen it, where you would have, let’s say, momentum system in Australia and a momentum system in the US, and you use different market regime filters. I wouldn’t agree with that. It doesn’t make sense to me. If it doesn’t work in one market, in theory, it shouldn’t work in the other, or to put it another way, your regime filter in one market should work in the other market. So I can take my mean reversion system from the US and throw it in the Australian market without any changes and that should, to some degree, show a reasonably good result. If I’ve got to go and change the regime filter to make it work, then possibly I’ve got to ask the question about the robustness of the strategy completely, so I wouldn’t agree with that. But again, it’s just another way to prove to myself the robustness of the strategy that if I take it from market A and stick it in market B, it doesn’t work, it doesn’t have to work perfectly, but doesn’t work, yes or no. If it completely dies, then you’ve probably got a problem that you need to be aware of.