Alternative Risk Premia strategies are based on a system of clearly defined rules, but there is less commonality in how these are applied than investors often recognise. GAM Systematic’s Dr Daniele Lamponi and Dr Anke Schorr explain that model dispersion arises because of decisions taken during all phases of developing a systematic harvesting algorithm. They illustrate why a combination of carefully considered investment rationales and extensive experience are ultimately key drivers in developing systematic investment strategies.
The academic credibility of Alternative Risk Premia (ARP) strategies is as important as their reliance on data and computer science to define their harvesting algorithms. In combination, these traits have created what we regard as an erroneous impression that ARP approaches are constructed from generic building blocks readily available to every investor in the market and that ARP, as an industry, is destined to become commoditised. This conclusion has been reinforced by the use of similar categorisations by asset class (foreign exchange, equities, interest rates, commodities, volatility) and by style (momentum, value, carry) – in other words, a shared risk premia phrasebook. We firmly believe that the opportunity set offered by ARP is essentially heterogeneous, at both the strategy and the aggregated portfolio level.
This lack of homogeneity is clearly demonstrated by the significantly different results, in terms of long-term performance, yielded across the ARP universe. For instance, the expected drawdown analysis favoured by GAM Systematic tends to be more counter-cyclical than methodologies based on risk parity and realised volatility, which tend to have higher weights in strategies that have recently shown strong performance. Even more importantly, however, we believe that the process of successfully harvesting single risk premia requires years of experience and constant attention to detail. For instance, it may appear easy to employ generic momentum models to extract risk premia from the foreign exchange markets. In reality, it is indeed complex to decide which specific contracts to consider and how to integrate them into an actual strategy.
Our long experience in the ARP space has clearly shown to us that an integrated process, whereby the ARP investment manager fully understands and controls the harvesting algorithms and their implementation, as well as the overall risk exposures within his or her portfolio, is an indispensable precondition for rewarding investment results. Properly managed, we believe the eventual outcome of such an approach will be outperformance.
In this article, we focus on the heterogeneity in ARP performance and illustrate possible sources of dispersion among risk premia harvesting algorithms. For illustrative purposes, we consider how a relatively straightforward (at first sight) foreign exchange trend model strategy could be designed. Such a model aims to capture the momentum risk premium in currency markets by dynamically switching between long and short exposure to the underlying currencies based on the observation of current versus historical prices. First discussed in equity markets (before FX rates were allowed to float in 1973), the momentum risk premium is well known and robust. It has been shown to exist across a broad range of markets and asset classes (Asness et al (2013)). Currency momentum has specifically been documented by Okunev and White (2003).
There are various explanations for the momentum effect and these are based on either behavioural finance or rational expectation assumptions. The former evidences herding1, investor over and under-reaction and the confirmation bias2, while the latter stresses exposure to global political and economic risks, and to hedging pressures (Filippou et al (2018)). Irrespective of the rationale behind it, the foreign exchange trend strategy is considered simple because its investable universe is relatively contained, while the individual instruments are very liquid. Transaction costs are low and there are no short-selling constraints. Common momentum signals have been widely employed as well as examined in academic literature (eg Guilleminot et al (2013)). As a result, we believe they can hardly be regarded as stemming from the special, obscure knowledge of certain opaque hedge funds. Finally, the pertinent data can be easily sourced and a basic algorithm requires limited programming skills and computational power.
This, or in fact any systematic investment strategy, can be represented by means of the five blocks or steps, as shown in Exhibit 1. Any systematic investment manager will, in principle, rely on such a structure.
Exhibit 1: Graphic representation of a systematic investment strategy
In particular, anyone designing an ARP model must make specific decisions on how to complete each of these boxes. These differing decisions will impact strategy characteristics, thus creating dispersion among investment managers harvesting the same risk premium. While outperformance is often explained by referencing the step involving the signals, it should be stressed that each step is important and will contribute to the final result. Exhibit 2 shows sample questions that can be addressed at each of the presented steps, as well as possible responses.Source: GAM Systematic. For illustrative purposes only.
Exhibit 2: Sample questions and possible responses to address at each step of an ARP strategy’s development
Source: GAM Systematic. For illustrative purposes only. Allocations and holdings are subject to change.
*The London fix, also known as the WM/Reuters fix, is compiled using actual transaction data from Thomson Reuters, ICAP-owned EBS and others. Spot and forward exchange rates are determined over a one-minute period, from 30 seconds before to 30 seconds after the time of the fix, which is generally 4.00pm in London. During this window, bid and offer rates from the order matching system and actual trades executed are captured. The final rates are then calculated by WM, a unit of State Street Corp
The specific choices made and parameters chosen will differentiate investment managers and product solutions. Some choices generate relatively minor differences, while others may have a substantial impact on overall exposure, risk profile and thus performance.
To illustrate these general points, we constructed a set of foreign exchange trend-following strategies. Exhibit 3 shows simulated Sharpe ratios obtained by running such models with different choices of parameters over almost two decades, from January 2000 to June 2019. These parameters comprise: the investable universe; a set of different momentum signals (for different time frames); and a set of methodologies to weight positions. For each run, we choose a set of contracts, pick one signal and a methodology to weight positions. The Sharpe ratios obtained by even this simple simulation differ widely, spanning a range from 0 to 0.6, averaging around 0.3, the expected level for such a strategy.
Exhibit 3: Sharpe ratios obtained by varying the parameters of a trend-following algorithm on currencies
Source: GAM Systematic. For illustrative purposes only. Simulations are run over the period January 2000 – June 2019.
The key questions are: how do we ensure choices are robust and optimally deliver on the model’s premise to harvest the respective risk premium? And what are the investment rationales, the expected features of the model and our overall modus operandi?
In-sample performance should clearly not be a driver of these choices, as this usually leads to what is known as ‘over-fitting’ (a modelling error that occurs when a function is too closely fitted to a limited set of data points) and poor out-of-sample performance. For the sake of strong in-sample performance, the number of parameters and choices to be made is often large. In most modelling frameworks, it is always possible to find the perfect back-test. As we do not focus on the past performance of investment strategies, what do we focus the modelling attention on instead? We aim to analyse the strategies’ risks and exposures, as well as to understand their behaviour and the performance drivers. Of course, having a long experience in harvesting risk premia is of great benefit in this process.
In-sample versus out-of-sample
When an investment strategy is tested, based on a certain choice of parameters, the period over which it is tested is called the in-sample period. The in-sample period might be the entire history of data available to run the strategy on or parts of that historical dataset. The resulting properties of the strategy, eg performance and risk features, are then called in-sample properties. The out-of-sample period is basically the dataset on which the investment strategy was not tested. In particular, since the strategy can only be tested on past data, the out-of-sample period contains all future data points. Any investment strategy seeks to deliver good risk-adjusted performance. Hence, risk strategies tend to display good performance referencing the in-sample period on which they are tested. However, this does not infer that the strategy will show similarly good performance over any (out-of-sample) period on which it has not been tested.
For illustrative purposes let us assume that the choices outlined in Exhibit 2 are followed to create an investment strategy. Namely G10 currencies are used as an investable universe (specifically, considering nine crosses against the US dollar), the daily London fix rates of the investable universe as data, a simple one-year price momentum (ie last year’s return) as a signal, an equal weighting (same gross exposure) scheme for the underlying contracts, and an ex-ante risk capped at 6%. The focus is on the exposures generated during the period of the back-test of the strategy. Key questions that need to be answered include: how is the model positioned and why? To do this, several analyses may be performed. Exhibit 4 shows the percentage exposure to the US dollar, either long or short. This latter is implicitly entailed in the strategy stemming from trading currency crosses against the US dollar. The value of 0% is the minimum and represents a portfolio where half the positions are long against the US dollar and the other half short, so that total US dollar exposure is zero3. The value of 100% represents a situation whereby all the positions have the same direction (long or short) against the US dollar.
Exhibit 4: Percentage exposure to US dollar, either long or short, which results from positions on crosses against US dollar
Source: GAM Systematic The value of 100% represents a portfolio where all the positions are long or short against the US dollar, while a value of 0% means that the total US dollar exposure is zero. For illustrative purposes only.
In the case of the model described so far, Exhibit 4 shows that significant US dollar exposure is predominant during the analysed period, which may suggest that the main performance driver is exposure to the US dollar. At this point, additional analysis would be needed to support our observation, such as for instance a principal component analysis, a mathematical procedure to analyse the variability present in a dataset.
This analysis helps find the relevant ‘dimensions’ that account for most of the variability in the dataset. For example, any combination of the nine included currency crosses could give rise to most of the variability. This means that we start out with a nine-dimensional problem. Principal component analysis, by means of the principal components, will help determine exactly what combination of the constituents brings about most of the variance. This is then the one relevant dimension which accounts for as much of the variance in the data as possible, with each subsequent component accounting for as much of the remaining variance as possible. Exhibit 5 shows the variance explained by the first two components over a one-year rolling window. It should be noted that the first component is much more significant than the second. It represents the main risk driver of the data sample. The more attentive reader may have already (correctly) guessed that this component can easily be interpreted as US dollar exposure. The second component results from alternating positions in commodity-related currencies, such as Australian and Canadian dollars, and safe haven currencies, such as the Japanese yen and Swiss franc. The first two components explain on average 74% of the variability of the dataset over the period considered.
Exhibit 5: Analysis of the variability present in the currency returns (principal component analysis)
Source: GAM Systematic The figure shows the percentage of variance explained by the first two components on a one-year rolling window over the period from January 2000 until June 2019. For illustrative purposes only.
This top-down analysis could then be complemented by a bottom-up approach whereby the single positions of the strategy during the period under consideration are reviewed to gain deeper insights into the strategy’s mechanics. The key questions here are: why the model is trading in or out of a position and how this trading contributes to portfolio risk and performance. For instance, a large part of the strategy’s performance could have been realised within a short time frame or due to a small number of events. For example, the capping of the Swiss franc in 2011 and the removal of this cap in 2015 deserve a close look. A simple trend-following model would have been positioned long the Swiss franc in 2011 and short in 2015, realising losses in both events. (Essentially, the explicit cap against the euro by the Swiss National Bank changed the dynamics of the Swiss franc.) Another illustrative point is provided by exposure to the British pound before and after the 2016 referendum on UK membership of the EU. The model was short the British pound as the run-up to the referendum saw the progressive devaluation of the British currency, while finally realising large gains in June after the referendum result.
We believe that understanding the risk and performance drivers of an investment strategy, as well as its underlying mechanics, helps to further improve this (or, indeed, any) model. If we are satisfied with US dollar exposure as the key implicit return driver of an FX trend model, the next step is to figure out the most efficient way to implement the strategy. Here, efficiency considerations include: transaction costs, liquidity and robustness. All previous choices should now be made as efficiently as possible. In the event that we are not satisfied and want to add diversification by decreasing the impact of US dollar exposure and augmenting the contribution of other performance drivers, we would need to act differently. Under such a scenario, expanding the investable universe and perhaps choosing a more complex methodology to weight the positions might appear a sensible option.
In conclusion, a combination of carefully considered investment rationales and extensive experience are key drivers in the development of harvesting algorithms. We believe that this combination ultimately leads to ever improving models.
1. Herding or herd behaviour represents the tendency for an individual to follow the actions of a group, whether those actions are rational or irrational (out of fear of being alone or missing out).
2. Confirmation bias is the tendency to search for, interpret, favour, and recall information in a way that confirms one's pre-existing beliefs or hypotheses.
3. As the investable universe is constituted by nine crosses the dollar exposure is always different from zero.
Asness, C., Moskowitz, T., & Pedersen, L. (2013). Value and Momentum Everywhere. The Journal of Finance, Vol. 68, pp. 929-985.
Filippou, I., Gozluklu, A. E., and Taylor, M. P. (2018). Global political risk and currency momentum. Journal of Financial and Quantitative Analysis, Vol. 53, pp. 2227-2259.
Grobys, K., Heinonen, J.-P., and Kolari, J. (2018). Return Dispersion Risk in FX and Global Equity Markets: Does It Explain Currency Momentum? International Review of Financial Analysis, Vol. 56, pp. 264-280.
Guilleminot, B., Ohana, J-J., and Ohana, S. (2014). Risk vs Trend Driven Global Tactical Asset Allocation. The Journal of Portfolio Management, Vol. 40, pp. 21-33.
Menkhoff, L., Sarno, L., Schmeling, M., and Schrimpf, A. (2012). Currency Momentum Strategies. Journal of Financial Economics, Vol. 106, pp. 660–684.
Okunev, J., D. White, (2003). Do Momentum-Based Strategies Still Work in Foreign Currency Markets? Journal of Financial and Quantitative Analysis, Vol. 38, pp. 425–447.
The information in this document is given for information purposes only and does not qualify as investment advice. Opinions and assessments contained in this document may change and reflect the point of view of GAM in the current economic environment. No liability shall be accepted for the accuracy and completeness of the information. Past performance is no indicator for the current or future development. Reference to a security is not a recommendation to buy or sell that security.
The examples provided in this material are being provided for illustrative purposes only. The examples provided were selected to assist the reader in better understanding the trading strategies presented and does not represent actual performance.
Important Information on Alternative Premia:
Alternative Premia strategies are speculative and are not suitable for all investors, nor do they represent a complete investment program. GAM Alternative Premia products are only available to qualified investors who are comfortable with the substantial risks associated with investing in Alternative Premia. Many of the investment programs are speculative and entail substantial risks. An investment in Alternative Premia strategies includes the risks inherent in an investment in securities, the use of leverage, short sales, options, futures, derivative instruments, investments in non-US securities and “junk” bonds. Investors should recognize that they will bear index based fees and expenses at the fund level, and indirectly, fees and expenses. In addition, the overall performance of Alternative Premia strategies products is dependent not only on the investment performance of individual indices, but also on the ability of a GAM investment manager to allocate assets amongst such indices on an ongoing basis. There can be no assurances that an investment strategy (hedging or otherwise) will be successful or that a manager will employ such strategies with respect to all or any portion of a portfolio. Alternative Premia strategies may be highly leveraged and the volatility of the price of their interests may be great. Investors could lose some or all of their investments. Investing in securities of foreign issuers involves special risks including currency rate fluctuations, political and economic instability, foreign taxes and different auditing and reporting standards. These risks are greater in emerging market countries. Leverage, including borrowing, may cause an underlying portfolio to be more volatile than if the portfolio had not been leveraged.