Description

Are you relying on a single trading signal to navigate the complexities of the foreign exchange market? If so, you might be missing out on the potential for enhanced profitability and reduced risk. In this engaging episode of Papers With Backtest: An Algorithmic Trading Journey, we dive deep into a groundbreaking 2019 research paper by Sonam Srivastava and colleagues, which unveils a multi-strategy approach to trading FX futures that could transform your trading game.

Join our hosts as they dissect the intricacies of combining various trading signals—including momentum, mean reversion, and carry trades—demonstrating how a diversified toolkit can significantly outperform reliance on a single indicator. This episode is packed with insights into the structured methodology employed in the paper, covering everything from instrument selection to signal creation and risk budgeting strategies. You'll gain a comprehensive understanding of how to craft a robust trading strategy that stands the test of market volatility.

Throughout the discussion, we meticulously analyze the performance of individual strategies, spotlighting standout performers like the long-term yield difference strategy while also addressing those that fell short. This thorough examination not only highlights the importance of strategy evaluation but also emphasizes the critical need for adaptability in algorithmic trading. The hosts reveal that the key to success lies in the synergy of multiple strategies, leading to significantly enhanced risk-adjusted returns.

As we explore different combination methods for these strategies, you'll discover how a diversified approach can mitigate risks and maximize returns, making a compelling case for traders to abandon the quest for a single optimal signal. Instead, you'll learn why building a robust toolkit of diverse indicators is essential for navigating the unpredictable waters of the FX market.

Concluding with a discussion on the importance of understanding market dynamics, our hosts underscore the potential for further research in this area, encouraging listeners to remain curious and innovative in their trading endeavors. Whether you are an experienced trader or just starting your journey, this episode of Papers With Backtest offers invaluable insights that can elevate your trading strategy to new heights.

Tune in and equip yourself with the knowledge to thrive in the ever-evolving landscape of algorithmic trading!

Hosted by Ausha. See ausha.co/privacy-policy for more information.

Transcription

• Speaker #0

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

• Speaker #1

  Indeed. The paper we're tackling today is a multi-strategy approach to trading foreign exchange futures. It's by Sonam Srivastava and colleagues from back in January 2019.

• Speaker #0

  Okay, FX futures. Yeah, that's a tricky market. What really caught my eye here is this idea, maybe instead of, you know, hunting for that one magic bullet indicator. Maybe combining different signals is actually the smarter way to go. So our mission today really is to get into the trading rules they looked at and critically what the back test showed. Does this multi-strategy thing actually work?

• Speaker #1

  That's exactly it. Yeah. The core argument is that blending indicators, you know, technical ones like momentum or mean reversion with something like FX carry. Well, it could potentially give you better, more consistent returns, more alpha, basically, than just trying to perfect a single approach.

• Speaker #0

  Right. So how did they actually test that? What was the process step by step?

• Speaker #1

  Well, they had a pretty structured methodology. First up, instrument selection. They focused on the big ones, the eight most liquid current month FX futures on the CME.

• Speaker #0

  Okay. Like the Euro Yen.

• Speaker #1

  Exactly. Australian dollar, British pound, Canadian dollar, Euro FX, Japanese Yen, Mexican peso, New Zealand dollar, and the Swiss franc. And they used a lot of data going back to

• Speaker #0

  1995. Wow.

• Speaker #1

  Okay. Yeah. And importantly, they used a T3 day rollover. That's just a way to handle the contract switching smoothly, avoid weird price jumps near expiry.

• Speaker #0

  Got it. Solid foundation. So what kinds of signals, what indicators did they build for these currencies?

• Speaker #1

  They built quite a few actually, and looked at both short-term, so three-month and long-term 12-month versions. First, there was interest rate carry.

• Speaker #0

  The classic carry trade logic.

• Speaker #1

  Pretty much. They measured it using the difference between long-term, that's 10-year, and short-term, one-year government bond yields. Specifically, the log difference. They call these short-term yield difference and long-term yield difference. Makes sense.

• Speaker #0

  Follow the yield.

• Speaker #1

  What else? Then standard technicals. Momentum, just looking at three-month and 12-month log returns. They label those short-term momentum and long-term momentum.

• Speaker #0

  Okay. Trend following.

• Speaker #1

  And the opposite, mean reversion, looking at returns relative to three-month and 12-month moving averages. Again, short-term mean reversion and long-term mean reversion.

• Speaker #0

  Standard toolkit stuff. And I thought you mentioned linking to other markets, too.

• Speaker #1

  Yes, exactly. They brought in equity momentum. The idea here is that some currencies, especially commodity or emerging market ones, tend to follow their local stock markets, right?

• Speaker #0

  Yeah, that linkage is known.

• Speaker #1

  So they calculated three-month and 12-month momentum for related indices, like the Aussie dollar linked to the ASX 200, the pound of the FTSE 100. You get the idea. Table one in the paper lists them all. These were the short-term equity momentum and long-term equity momentum signals.

• Speaker #0

  Interesting. Connecting the dots across assets. What about commodities themselves?

• Speaker #1

  Yep. Covered that too. Commodity momentum. They looked at the SPGSEI index, Brent crude, gold, and an agriculture index. Again, both short-term, three-month, and long-term, 12-month versions. Short-term commodity momentum, long-term commodity momentum. Oh, okay. And one last one. Realized volatility. They created short and long term indicators based on how choppy the price action of these currency had been historically. Short term volatility, long term volatility.

• Speaker #0

  Right. So that's a pretty wide range of signals. How did they compare them fairly? They must operate on totally different scales.

• Speaker #1

  Good point. That's where normalization came in. They squashed all the raw indicator values into a standard range. Minus 0.5 to plus 0.5. Ah, OK. They use a walk forward percentile method. Basically, it adjusts the scaling based on recent history. So it adapts a bit. This meant they could apply consistent position sizing rules later.

• Speaker #0

  Right. You need that common scale. And the sizing, how much did they bet on each signal?

• Speaker #1

  They used risk budgeting. The goal was to aim for 10% annualized volatility for each individual indicator strategy. And interestingly, they allowed negative allocations.

• Speaker #0

  Meaning they could go short based on a signal.

• Speaker #1

  Exactly. Not just long signals.

• Speaker #0

  Okay. So now we have all these individual strategies normalized with risk targets. This is the core of it, right? How did they combine them?

• Speaker #1

  Yes, this was a major part of their investigation. They tried a whole bunch of different combination methods. Simple stuff like equal weighting. Just give each strategy the same slice of the pie. Okay,

• Speaker #0

  easy enough.

• Speaker #1

  And more complex things like risk parity, where you allocate so each strategy contributes the same amount of risk. Or weighting by Sharpe ratio, giving more weight to historically better performers.

• Speaker #0

  With or without considering correlation?

• Speaker #1

  Both. They had versions proportional to Sharpe. and a correlation-aware sharp weighting. Plus, optimization methods like trying to maximize diversification, basically, picking weights to make the combined strategies as uncorrelated as possible.

• Speaker #0

  Trying to smooth out the ride.

• Speaker #1

  Exactly. And a couple of others aimed at maximizing sharp directly, or even maximizing the 10th percentile of the rolling one-year sharp. A more conservative optimization.

• Speaker #0

  Okay, a lot of ways to mix and match. So first things first, how did those individual strategies perform on their own? Any standouts?

• Speaker #1

  Yeah, if you look at their table three, the results are pretty clear. The single best performer, by quite a margin actually, was the long term yield difference strategy. Highest return, highest sharp.

• Speaker #0

  Huh, just following the long term rate differentials. Interesting.

• Speaker #1

  Short term commodity momentum and short term yield difference also did okay on their own.

• Speaker #0

  And the losers, any single strategies that just didn't work well?

• Speaker #1

  Well, according to their back tests, yeah. Short-term mean reversion, long-term volatility, and long-term equity momentum. They didn't really deliver strong risk-adjusted returns individually during this period.

• Speaker #0

  Good to know. Okay, now for the main event. What happened when they started combining these using those different methods, starting with the simple uniform weights across all currencies?

• Speaker #1

  Right, so in this first combination approach, the same weighting rule applied to all eight currencies. Looking at Table 4 and Figure 3, the results were, well, quite encouraging for the multi-strategy idea. The best combination methods significantly outperformed the best single strategy. The top performer was actually the one optimizing for the 10th percentile of the rolling sharp.

• Speaker #0

  Ah, the more conservative optimization paid off.

• Speaker #1

  It seemed so. Volatility-scaled negative correlation and maximum diversification also did pretty well. It really showed that combining things added value beyond just picking the best individual signal.

• Speaker #0

  That's a really key finding. But you mentioned some combination methods weren't great.

• Speaker #1

  Yes, they specifically called out that simply optimizing to maximize the overall Sharpe ratio didn't perform that well.

• Speaker #0

  Ah, the classic overfitting trap.

• Speaker #1

  Looks like it. It suggests just chasing the highest historical Sharpe can lead you down a bad path. Focusing on consistency, like that tenth percentile method or diversification, seems more robust.

• Speaker #0

  That makes a lot of sense. Don't just fit the noise. OK, so then they took it even further, right? Allowing different indicators for different currencies.

• Speaker #1

  Exactly. The product X indicator combinations. This was more complex. They basically treated each currency indicator pair like Aussie dollar long term yield difference as its own mini strategy.

• Speaker #0

  Whoa, that massively increases the number of things to combine.

• Speaker #1

  It does. They did filter out some of the really poorly performing individual pairs first, heuristically, before combining the rest. Table five shows some examples like. Long-term yield difference was useful for several currencies, but maybe equity momentum only for specific ones.

• Speaker #0

  And the results? Did this more granular approach boost performance again?

• Speaker #1

  It really did. Table 6 and Figure 6 show another jump in performance. Again, maximizing diversification and maximizing the 10th percentile of sharp were standouts.

• Speaker #0

  How much better are we talking?

• Speaker #1

  Significantly better. The sharp ratio for the best combination here was roughly 60% higher. than the sharp of the best single indicator strategy.

• Speaker #0

  Wow. A 60% improvement is huge. That really drives home the point, doesn't it?

• Speaker #1

  Absolutely. It strongly suggests that different factors matter more for different currencies. A one-size-fits-all approach, even in combination, isn't necessarily optimal. Tailoring the indicator mix per currency pair paid off, at least in this backtest.

• Speaker #0

  So if we boil it all down, what's the big takeaway for someone trading FX systematically?

• Speaker #1

  I think the core message is pretty clear. Don't just look for the single best signal. A multi-strategy approach where you thoughtfully combine diverse indicators carry, momentum, vol-vol, maybe even cross asset links using robust methods, well, it has the potential to seriously improve your risk-adjusted returns.

• Speaker #0

  And robust methods seems key there, focusing on diversification or consistency, not just chasing peak historical sharp.

• Speaker #1

  Precisely. Methods like maximizing diversification or optimizing for that lower percentile of rolling sharp. seem particularly effective here. It's about building a diversified portfolio of strategies, not relying on one potentially fragile edge.

• Speaker #0

  That definitely shifts the perspective. It's less about finding the silver bullet and more about building a robust toolkit.

• Speaker #1

  Exactly. Thinking about how different market dynamics interact and capturing them systematically. It's definitely food for thought. The researchers even mentioned looking deeper into why certain factors work for certain currencies, maybe bringing in macro data too. Lots more to explore there.

• Speaker #0

  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.