Transcription

• Speaker #0

  Hello, welcome back to Papers with Backtest podcast. Today we dive into another algo trading research paper.

• Speaker #1

  Yes, this one looks at trend following in futures.

• Speaker #0

  We're going to explore a study that really gets into the details of trading rules and the actual historical results, you know, from backtesting these trend following strategies.

• Speaker #1

  Specifically for the futures market.

• Speaker #0

  Exactly. Get ready to see how some, well, surprisingly simple ideas have actually held up when you put them to the historical test.

• Speaker #1

  Right. And our goal here, you know, in this deep dive is to pull out the key learnings. How do these strategies work? What's their track record look like? Profits, losses, the whole picture.

• Speaker #0

  And what you, the listener, can maybe take away for thinking about your own approach to algorithmic trading, specifically in futures.

• Speaker #1

  Actionable insights,

• Speaker #0

  hopefully. Okay. So the paper kicks off by laying out two pretty fundamental trend following strategies. First up, one most folks in this space will know. The moving average crossover. Yeah,

• Speaker #1

  that's a classic one.

• Speaker #0

  Can you break down the sort of core mechanics? How does the research describe it?

• Speaker #1

  Sure. So the basic idea is you use two moving averages. One's faster, reacts more quickly to recent prices. Okay. And the other's slower, smoother, uses daily price data. And the rule is buy when the faster average crosses above the slower one.

• Speaker #0

  Right, like short-term momentum is picking up.

• Speaker #1

  Exactly. Yeah. And conversely, it's a sell signal or go short. when that faster average dips back below the slower one. Okay. And the key thing here is with this setup, you're always in the market, either long or short.

• Speaker #0

  Right. Always got skin in the game, so to speak. Okay. The second strategy is a breakout strategy. Now, this sounds like it operates a bit differently.

• Speaker #1

  It does. Yeah. It's looking for prices to kind of break out of recent ranges.

• Speaker #0

  How does that work? The entry rule.

• Speaker #1

  So for going long, the rule is buy tomorrow. If today's closing price is higher than or equal to the highest close seen in the past, say, 50 days, that's the look-back period.

• Speaker #0

  Okay, 50 days. Makes sense.

• Speaker #1

  And for a short position, it's triggered if today's close is lower than or equal to the lowest close over that same 50 days.

• Speaker #0

  Got it. And getting out.

• Speaker #1

  Exits are also based on recent action. You sell a long position if the close hits the lowest point of the last 25 days.

• Speaker #0

  25 days for the exit?

• Speaker #1

  Yep. And you cover a short position. if the price hits a 25-day high. And importantly, like the crossover, it looks at daily closing prices. But the trades happen the day after the signal. So unlike the crossover, this one isn't always holding a position.

• Speaker #0

  Ah, okay. So it can be flat sometimes. Interesting difference. All right, we've got these two strategies. Now, the crucial part, how did they actually do when the paper ran them through historical data?

• Speaker #1

  Right. the backtest results. Yeah.

• Speaker #0

  What were the key numbers, the returns, the drawdowns, and how did they stack up against, you know, standard market benchmarks?

• Speaker #1

  So the backtest covered a decent stretch, January 1990 through December 2011. It gives a good long-term picture. For both strategies, the paper really focuses on compounded annual return, your average yearly gain reinvested. Then the worst drawdown, that's the biggest peak to trough loss you'd have seen.

• Speaker #0

  Always important to know that one.

• Speaker #1

  Absolutely. And the ratio of that drawdown to the annual return, kind of a risk-reward measure.

• Speaker #0

  Okay. And the benchmarks?

• Speaker #1

  Well, they compared against the MSCI World Index, you know, broad global stock. That had an annualized return of about 4.7%, but a really hefty max drawdown, 57.5%.

• Speaker #0

  Oh, coach. Yeah, that's big.

• Speaker #1

  And then there's the Barclay BPOP 50 Index. That tracks managed futures traders. It showed 7.3% annualized return, but with a much lower max drawdown, just over 13%.

• Speaker #0

  Wow. Quite a difference there.

• Speaker #1

  Yeah. And the paper also mentions performance against a couple of specific futures programs, Milburn and Dunn. But the exact numbers aren't in this particular table we're looking at.

• Speaker #0

  OK. But the main point seems to be even these simple strategies, well, they potentially outperform traditional stocks during that simulation period.

• Speaker #1

  And importantly, maybe with less gut-wrenching drops in value, judging by those benchmark drawdowns.

• Speaker #0

  Yeah. That BTOB 50 comparison is interesting, too. Now, here's where it gets really interesting for tweaking things. The paper looked at that breakout strategy and fiddled with the look-back periods, those 25 and 50-day windows.

• Speaker #1

  The parameters.

• Speaker #0

  What happened when they changed those numbers?

• Speaker #1

  Yeah, they specifically tested look-backs of 25, 50, and 100 days for the entry signal on the breakout. And what they found, looking at Table 4.3, was actually a good degree of stability.

• Speaker #0

  Stability, meaning?

• Speaker #1

  Meaning all three versions, 25, 50, 100 days, produced pretty strong results in the back test. It suggests the core idea of the breakout wasn't super sensitive to that exact number.

• Speaker #0

  Ah, so the concept itself seems robust. That's good to know. You're not just curve-fitting one perfect number.

• Speaker #1

  Exactly. It hints that the general principle breaking out of a recent high or low was the key driver. More than the precise length of that look back window.

• Speaker #0

  That's really useful. Okay, now, a common headache with trend following is getting whipsawed, right? Market chops around, you get false signals. Oh, yeah. The paper talks about adding a trend filter to maybe help with that. What's the thinking there, and did it actually improve things?

• Speaker #1

  Right. The idea behind adding a trend filter is basically to try and avoid taking trades that are fighting the bigger underlying market direction. Like, don't buy in a downtrend, don't shorten an uptrend.

• Speaker #0

  Take a sanity check?

• Speaker #1

  Sort of, yeah. A second opinion before you jump in. The paper mentions using a slower moving average as this filter. So the rule becomes, like, only take buy signals if the price is also above this slow average.

• Speaker #0

  And only short signals if the price is below it.

• Speaker #1

  Precisely. And when they tested the 25 by 50 breakout strategy with this filter versus without it, the results in Table 4.4 showed a clear improvement.

• Speaker #0

  Improvement how?

• Speaker #1

  Higher compounded annual return 22.4% with the filter versus 19.4% without. Nice. Lower worst drawdown to drop from about negative 32 percent down to negative 26 percent.

• Speaker #0

  Even better risk reduction.

• Speaker #1

  And even a slightly higher percentage of profitable months. So, yeah, seemed like a worthwhile addition in the simulation.

• Speaker #0

  OK, so adding that filter seemed to boost returns and control risk better. Interesting. The paper also looked at tweaking the stop loss for the breakout strategy. The initial rule was that 25 day low or high exit. Maybe that has downsides.

• Speaker #1

  It can, yeah. The paper points out that if you're in a really strong trend, waiting for the price to pull all the way back to a 25-day low if you're long, or a 25-day high if you're short, well, that could mean giving back a lot of your profit if the trend just pauses or corrects a bit.

• Speaker #0

  Right, the trend might still be intact, but you get stopped out too late.

• Speaker #1

  Exactly. So to try and address this, they experimented with a volatility-based stop mechanism.

• Speaker #0

  Volatility-based? How does that work?

• Speaker #1

  Well, the paper doesn't lay out the exact formula here, but the concept is to use a measure of recent price movement volatility to set your stop distance.

• Speaker #0

  Ah, so tighter stops when things are calm, wider stops when they're choppy.

• Speaker #1

  That's the general idea, yeah. Give the trade room to breathe when volatility is high, but lock in profits or cut losses quicker when things calm down. The goal is maybe a bit more predictability in the exit.

• Speaker #0

  And the results, did that help?

• Speaker #1

  It was a trade-off. Looking at table 4.5. The backtest showed a decrease in overall profitability, but it did significantly cut the maximum drawdown again down to 20 percent. So you ended up with about an 18 percent compounded annual return, but with less risk measured by that max drawdown.

• Speaker #0

  Interesting. So you sacrificed some potential return for tighter risk control, a classic tradeoff.

• Speaker #1

  Pretty much. Yeah. And then they also touched briefly on position sizing, specifically using a point one percent risk factor per trade. What was the main takeaway there?

• Speaker #0

  Right. Figure 4.16 and 4.17 show this. By using that consistent risk factor, they could show a year-by-year performance comparison against the BTOP 50 index.

• Speaker #1

  Anything jump out?

• Speaker #0

  The big one was the significant outperformance in 2008.

• Speaker #1

  Ah, the financial crisis year.

• Speaker #0

  Exactly. It really highlights that potential strength of trend following the ability to possibly profit when other markets are under major stress.

• Speaker #1

  Yeah, that crisis alpha potential. Okay, let's shift gears a bit to Chapter 5. The paper digs into the distribution of the individual trades. What did the result of all those single bets look like?

• Speaker #0

  Figure 5.1 paints a really interesting picture. It shows that the most frequent outcome for any given trade was actually a small loss.

• Speaker #1

  A small loss, not a small win.

• Speaker #0

  Nope, a small loss, typically in the range of maybe 0.5% to 0.75% of the portfolio. And this is really characteristic of many trend following systems.

• Speaker #1

  Right. Lots of small paper cuts.

• Speaker #0

  Kind of. You tend to get many small losses, a decent number of small or medium winners, but the real profit engine that comes from the less frequent but much, much larger winning trades.

• Speaker #1

  The outliers, the fat tails of the distribution.

• Speaker #0

  Exactly. Those big winners are what ultimately make the whole diversified approach profitable over time.

• Speaker #1

  So it's definitely not about having a high win rate. It's about making sure your winners are significantly bigger than your losers on average.

• Speaker #0

  Precisely. OK. The paper also split the performance out. What if you only traded long? What if you only traded short?

• Speaker #1

  Oh, yeah. Table 5.2 shows this. And the results were, well, quite revealing.

• Speaker #0

  So, so.

• Speaker #1

  The short-only version of the core strategy, it actually performed pretty poorly on its own. Low overall return, something like 2.5% compounded in a really long time underwater. The drawdown relative to the return was huge, almost 13 years worth.

• Speaker #0

  Yikes. So shorting alone wasn't great. What about long-only?

• Speaker #1

  Long-only did much better. about 15.5% compounded return with a 14.7% drawdown, much healthier looking.

• Speaker #0

  Okay, so long did well, short did poorly. But the combined strategy did best overall, right?

• Speaker #1

  Yes. And that's the crucial point. Even though the short side look weak standalone, combining it with the long side provided really valuable diversification.

• Speaker #0

  Diversification happened.

• Speaker #1

  Especially during equity bear markets. When stocks are tanking, having those short positions, maybe in other asset classes, can really help cushion the blow to the overall portfolio.

• Speaker #0

  Got it. So the short side's value isn't just its standalone profit, but its role as a hedge or diversifier when things get ugly elsewhere.

• Speaker #1

  Exactly. Makes the whole system more robust.

• Speaker #0

  Makes sense. The paper also gave a quick look at performance across different sectors and futures. Any differences there?

• Speaker #1

  Yeah. Table 5.3 gives a snapshot of average yearly returns by sector. Things like agricultural and non-agricultural commodities, they generally showed positive returns for both long and short sides.

• Speaker #0

  Okay.

• Speaker #1

  Currencies tended to do a bit better on the long side. Interestingly, equity futures, they showed lower overall returns in this back test.

• Speaker #0

  Lower? Why might that be?

• Speaker #1

  Well, the paper speculates it could be factors like increased electronic trading, more HFT activity maybe, and possibly higher correlations between global stock markets, making trends harder to catch.

• Speaker #0

  Plausible reasons. And rates.

• Speaker #1

  Interest rate futures also had their own distinct performance patterns, as you'd expect.

• Speaker #0

  Right. OK, so moving towards wrapping up, the paper gets into some really practical stuff in Chapter 9. One big one is the need for a decent amount of capital, right? A sufficient asset base. Why is that so critical for futures trend trading?

• Speaker #1

  Yeah, it's really important. It comes down to futures contract sizes and margin requirements. You need enough capital to actually trade a properly diversified portfolio without taking on insane risk on any single position.

• Speaker #0

  Can you give an example?

• Speaker #1

  Sure. The paper uses an example. Imagine you have a $150,000 account. You're using a volatility-based position sizing method, like the ATR-based one. For a market like, say, LiveTattle, The formula might tell you to risk a certain dollar amount, which translates to buying, I don't know, 0.3 contracts.

• Speaker #0

  But you can't buy 0.3 contracts.

• Speaker #1

  Exactly. You can only trade whole contracts. So if your account size isn't large enough relative to the contract size and volatility, your position sizing formula might tell you to take a trade size that's physically impossible. You either have to skip the trade or take on much more risk by rounding up to a full contract.

• Speaker #0

  Ah, so for smaller accounts, getting that that smooth diversification across many markets, like in the . backtest might just not be feasible.

• Speaker #1

  It can be a real challenge, yes. You might have to concentrate in fewer markets or accept deviations from the ideal position sizing.

• Speaker #0

  Makes perfect sense. Another really practical point they stress is about starting to trade live. How should you approach that initial jump from backtest to real money?

• Speaker #1

  This is key. The paper strongly advises that when you go live, you need to enter all the positions that your strategy signals at that moment.

• Speaker #0

  Not just wait for new signals to come along.

• Speaker #1

  Right. Don't just start flat and only take the next buy or sell signal. The backtest results assume the strategy was fully invested according to its rules throughout the period.

• Speaker #0

  So if the backtest shows you'd currently be long gold, short bonds, and flat crude...

• Speaker #1

  Then on day one of live trading, you should ideally establish those exact positions. If you just wait for the next signal, maybe a signal to buy crude, you're already deviating. You're making a discretionary choice. Not to hold the golden bond positions the model says you should have.

• Speaker #0

  And that means your live performance could look very different from the back test because you didn't replicate the starting portfolio conditions.

• Speaker #1

  Precisely. You need consistency right from the get-go. Stick to the rules fully.

• Speaker #0

  Got it. So consistency, adhering to the rules from day one, is vital if you want a realistic shot at matching those simulated results. Okay, this has been really insightful. A good look under the hood of these trend-following strategies and futures.

• Speaker #1

  Absolutely. I think the key takeaways are even relatively simple trend rules can show strong long term potential in futures markets. That seems clear from the back test. But and it's a big but you absolutely need to understand the results properly. That includes the drawdowns. It'll happen and understand where the profits come from. The mix of long versus short different sectors.

• Speaker #0

  Right. It's not magic. It's about understanding the characteristics.

• Speaker #1

  And then there are the practical things. Having enough capital, like we just discussed, and actually implementing the strategy faithfully when you go live. Those are critical.

• Speaker #0

  Great summary. Thanks. My pleasure. Thank you for tuning in to Papers with Backtests podcast. We hope today's episode gave you useful insights. Join us next time as we break down more research. And for more papers and backtests, find us at https.paperswithbacktests.com. Happy trading.