The Need for Speed in Trend-Following Strategies

Why reactive trend-following strategies are the wingman of traditional investors.

“I Feel the Need, The Need For Speed”

At Man AHL1 we empathise with what is probably Maverick’s most famous quote in 1986’s “Top Gun”. Following trends quickly and being responsive to emergent (or dissipating) changes in market directions, is a design goal for all of our trend-following strategies. As we explain in this note, faster trend systems have attractive risk-management properties and are more complementary to traditional investments than slower systems.

What Is ‘Speed’ In Trend-Following?

Academic studies have shown that trends exist in markets over different time horizons, with some persisting for a few days or weeks, and others running for several months (Moskowitz et al., 2011). By ‘speed’ we mean trend-length sensitivity; ‘fast’ and ‘slow’ trend systems focus on capturing the short- and long- end of this spectrum respectively. While there are a variety of algorithms that can be used to identify trends, in this note we will investigate performance characteristics of a suite of double exponentially weighted moving-average crossover (‘MAC’) models, as per Table 1. These, or variations thereof, have been in use at Man AHL for around three decades and still represent the model with the greatest risk allocation in Man AHL’s trend-following strategies. The choice of trading speeds is chosen to both span the range of trends we are seeking to capture, and minimise correlation between the models (see Appendix A). For example, the correlation between the fastest and slowest speed is 0.17.

Table 1. Reference for Noted Trading Speeds

Problems loading this infographic? - Please click here

Source: Man Group Database. Please see the important information linked at the end of this document for additional information on hypothetical results.

To determine a model’s performance characteristics, we backtest it from 1995 through to 2022 across the 50 most liquid futures and FX forward markets. Risk allocations are split equally across asset classes: equity (25%), fixed income (25%), FX (25%) and commodities (25%). Individual markets are volatility scaled such that each has equal risk weight within an asset class. The list of markets and risk allocations are shown in Appendix B, and high-level results are shown in Table 2.

Table 2. High-Level Statistics of Trend-Following Speeds

Problems loading this infographic? - Please click here

Date range, 1 January 1995 to 31 August 2022. Skewness calculated using montly overlapping returns.
Source: Man Group Database. Please see the important information linked at the end of this document for additional information on hypothetical results.

As would be expected, turnover decreases with slower speeds which, as we will see, has implications in the real world via transaction costs. Reassuringly, Sharpe ratios are all significantly positive. Skewness is positive for almost all speeds, but is clearly more so for fast strategies.

Perhaps the most interesting aspect of Table 2 is the apparent trade-off between Sharpe ratio and skewness; risk-adjusted returns increase with slower speed, but risk-management properties, via skewness here, deteriorate. The intuition here is that faster models truncate losses quickly when a trend reverses, cutting off that left tail, while still allowing profits to run. In fact, this is a key property of univariate trend-following strategies in general, rather than the dynamics of markets, as demonstrated by Martin & Zou, 2012.

Slower trend models also have modestly higher correlation to traditional asset classes, which might be expected since we expect traditional assets to increase in price over the long term because of their embedded risk premia, which may be captured by the slowest measures of trend.

The Need for Speed

Our analysis thus far has shown that returns from our MAC models at different speeds are positive in the long term and are lowly correlated to each other. A systematic mindset says that this diversification should be captured by trading all the speeds, easily afforded by automation, thereby increasing risk-adjusted returns and, with the judicious use of leverage, returns themselves.

But what weights should we allocate to each model speed? As we have found, faster trading potentially delivers greater risk-management properties, but at some cost to Sharpe ratio. However, this can still be particularly advantageous when combined with traditional investments because of the consistent low or negative correlation over the long term.

At Man AHL, we find a persuasive argument for having proportionate weights to fast trend models through the analysis of ‘Crisis Alpha’. This is the term used to describe trend-following’s historically observed property of performing well in risk-off environments. In Figure 1 we plot the performance of each of our speeds by S&P 500 return quintile – around one month holding period on the top, and around three months on the lower plot.

Figure 1. Performance By Speed During Equity Return Quintiles

Problems loading this infographic? - Please click here

Date range: 1 January 1995 to 31 August 2022. Each model speed is scaled to 10% annualised volatility (ex post).
Source: Man Group database, Bloomberg.

The average annualised return for both time horizons studies – one month and three months – generally improves as speed of trading decreases, as we found earlier. However, convexity and performance when the S&P 500 is in its worst quintile, our “Crisis Alpha”, increases as the speed is increased.

We can further investigate this effect by examining the asset class performance by speed during the worst S&P 500 return quintile (across 21- and 65-day returns), shown in Figure 2 below.

First, we show that regardless of speed, trend systems generate their “Crisis Alpha” from gains in all asset classes, not just equities. Prior to 2022, these have been dominated by long bond positions as crises in equities have typically led to a flight-to-quality of bonds effect. Hamill et al., 2016 and Robertson, 2022, suggest this is potentially an artifact of generally rising fixed income prices in the backtest period.

Second, positive equity attributions are typically a feature of faster trend models. The slowest trend speed cannot shift to a short position over a one- or three-month horizon. To us, this is crucial given that investors may often review performance, and therefore investments, on a monthly or quarterly basis. This was of great significance during the short-lived COVID-led equity rout in Q1 2020. If “Crisis Alpha” is a desired outcome of an allocation to trend following, then a responsive trend system is key to ensure that outcome.

Figure 2. Performance by Speed by Asset Class During Worst Equity Return Quintile

Problems loading this infographic? - Please click here

Date range: 1 January 1995 to 31 August 2022. Each model speed is scaled to 10% annualised volatility (ex post).
Source: Man Group database, Bloomberg.

The Need for Speed Execution

As always, the real world has the potential to get in the way. Transaction costs impact faster trading speeds disproportionately because of the higher turnover (Table 1) and therefore more frequent crossing of the bid-offer spread. Using Man AHL’s trading cost models, built using three decades of experience trading trend-following strategies at scale, we calculate Table 3, which extends Table 2.

Table 3. Performance And Skewness of Trend Speeds, Before And After Costs

Problems loading this infographic? - Please click here

Date range, 1 January 1995 to 31 August 2022. Simulated after cost results are based on portfolio running USD 5bn.
Source: Man Group Database. Please see the important information linked at the end of this document for additional information on hypothetical results

The Sharpe results are somewhat intuitive; risk-adjusted returns after costs are materially lower for faster speeds over the long-term. Interestingly, skewness properties remain largely intact, and unaffected by the addition of costs.

Figure 3 is the with-costs analog of Figure 1, illustrating the effect of transaction costs on “Crisis Alpha”. As expected, average returns at faster speeds are impacted more once transaction costs are included but, consistent with our skewness findings of Table 3, performance during equity weakness is still best with faster trading.

Figure 3. Performance by Speed During Equity Return Quintiles (Including Trading Costs)

Problems loading this infographic? - Please click here

Date range: 1 January 1995 to 31 August 2022. Each model speed is scaled to 10% annualised volatility (ex post).
Source: Man Group database, Bloomberg.

It stands to reason, therefore, that efficient execution is the gatekeeper to being able to trade fast. Maverick may have felt the need for speed, but he needs his F-14 to get there. Man AHL’s F-14 is a purpose-built execution platform, with two cornerstones. First, algorithms are tuned to Man AHL’s style of trading, taking advantage of the fact that trends last at least several days, and hence trades can mostly be dripped slowly and passively into the market, and only aggress when deemed necessary. Second, flow is disguised to minimise the predictability of trades and hence reduce the negative impact of high-frequency traders. Broadly, we find that Man AHL reduces transaction costs by a factor of two over bank algorithms.

Diversification in a Traditional Portfolio

We have shown that there is no perfect trend speed; slow speeds can indeed deliver good long-term performance but lack the attractive risk-mitigating properties of faster speeds. However, to our knowledge, very few investors own solely trend-following strategies. Instead, they tend to be used as part of a portfolio. If the aim of the trend-following allocation is to boost the defensive properties of a portfolio, then perhaps a more responsive system, allocating more to fast trend models, may suit best in our view. We explore this below by comparing the drawdown profile of a traditional 60/40 portfolio2 (“60/40 portfolio”) combined with various trend strategies. In order to emphasise any drawdown impact, we choose an equal allocation between the two components.

The 60/40 portfolio is combined with three different trend strategies. As we discussed earlier, the low correlation of the returns of different trend speeds suggests their combined use in a diversified trend-following system. Therefore the first trend strategy, dubbed “MOM Equal Blend” combines all five trend speeds by taking the equally weighted sum of our individual trend speeds. The second trend strategy is slower by choice. Dubbed “MOM Slow Blend”, it combines the slowest two trend speeds only. For both of the above trend strategies we introduce a speed diversification scaling factor to address the resulting lower portfolio volatility that arises when combining diversifying speeds. Finally, we also compare “MOM V.Slow” which was our slowest trend speed highlighted earlier in this paper. All trend strategies are adjusted to reflect 10% return volatility before being combined with the 60/40 portfolio.

Figure 4 shows the drawdown chart of each combined portfolio as well as the 60/40 portfolio without an allocation to a trend strategy, alongside values at key drawdown episodes. Here, drawdowns are defined as peak-to-current returns at each point in time. As expected, all combinations with a trend strategy deliver some degree of risk mitigation compared to the traditional portfolio. Moreover, the degree of downside mitigation typically improves with greater allocation to faster speeds.

The results suggest that, just like Maverick, investors in trend-following, particularly those seeking defensive properties, should feel the need for speed.

Figure 4. Performance by Speed During Equity Return Quintiles (Including Trading Costs)

Problems loading this infographic? - Please click here

Problems loading this infographic? - Please click here

Date range: 1 January 1995 to 31 August 2022. 60/40: Represented by 60% allocation to the MSCI World Index and 40% allocation to the Barclays Global Aggregate Bond Index. The trend portfolios have been scaled to 10% annualised volatility (ex post) prior to being combined with the 60/40 portfolio. Source: Man Group database, Bloomberg. Please see the important information linked at the end of this document for additional information on hypothetical results.

Note: Appendices are available in the PDF of this document.

 

Bibliography

Hamill, C., S. Rattray, and O. Van Hemert. (2016). “Trend Following: Equity and Bond Crisis Alpha”. Available at SSRN: https://ssrn.com/abstract=2831926

Martin, R. and D. Zou. (2012), “Momentum trading: ‘skews me”, Risk, 25(8), 52-57.

Moskowitz, T., Y. Ooi, and L. Pedersen (2012), “Time series momentum”, Journal of Financial Economics, 104(2), 228–250.

Robertson, G. (2022), “Gaining Momentum: Where Next for Trend-Following?”, Man Institute, Available at: https://www.man.com/maninstitute/gaining-momentum-trend

 

1. The author is grateful for contributions from Martin Luk, Graham Robertson, Matthew Sargaison and Otto van Hemert.
2. The traditional 60/40 portfolio is based on 60% MSCI World Net Total Return and 40% Barclays Capital Global Aggregate Bond index rebalanced monthly.

User Country: United States (237)
User Language: en-gb
User Role: Public (Guest) (1)
User Access Groups:
Node Access Groups: 1