A shorter version of this article was originally posted on  For additional behind the scenes insight, see commentary provided at the end of this article.

By Mark Melin on June 13, 2017 1:28 pm in Business Edit

Can analysts predict returns for alternative investment strategies and used in a hedge fund selection methodology? European researchers studying UCITS fund returns think returns are predictable. Several caveats and many algorithmic formulas that add layers of complexity later, the result, the researchers say, is yes, returns can be predicted and they can be used in a hedge fund selection methodology, but there are environmental factors to consider. Regardless of the report conclusion, it exposes the hot topic of beta market environment research in alternative investments.

Predictive variables” include some factors, such as a moving average cross trend benchmark methodology, that are backward looking and researchers say are not predictive, based on their low individual trade win percentage and dependence on large win size. Researchers used the moving average cross methodology of determining how momentum might be playing into overall market movement mechanics. 

Market environments can be understood relative to an algorithmic strategy, but can researchers predict returns?

In some respects, the academic work of researchers Michael Busack, Wolfgang Drobetz and Jan Tille, European practitioners and academics, are touching on a cutting edge topic inside the world of quantitative finance: Can researchers model market environments and predict returns to use this in a selection process?

That is a loaded question which the authors, sticking their academic and professional necks out, say is possible.

Busack and Tille are portfolio building practitioners for Hamburg, Germany-based Absolut Return while Drobetz is on the faculty at Hamburg University, and recently published their work.

The report starts by pointing to research that supports the generally accepted claim that a stock sector or alternative investing strategy can be modeled to a market environment:

There is also empirical evidence that hedge fund returns are predictable as a result of their exposure to macroeconomic and market risk factors (Amenc et al., 2003; Wegener et al., 2010; Bali et al., 2011; Drobetz et al., 2012; Avramov et al., 2013; and Panopoulou and Vrontos, 2015; among others).

But next the report moves into new territory, predicting returns.

NonCorrelated investing, uncorrelated returns, beta market environment, market cycle analysis

The list of predictive returns is not entirely predictive

Modeling a stock selection strategy to a macro market cycle is not uncommon, but predicting returns involves another component

Many managing alternative investment portfolios and mainstream equity analysts are commonly found mapping a stock sector or particular strategy to a market environment. A recent Moody’s report published in ValueWalk, for instance, is but one example of modeling a macro market cycle ”rising interest rates” and evaluating how this impacts different stocks. In a rising interest rate market environment, it is common in fact expected to see analysts note which business models favor rising rates “with brokerages and certain bank types commonly considered relative to how their business model models through different interest rate environments. There is software that tracks various equity market cycles and maps this to specific stock sector selection from Optimal Asset Management, for instance, that helps advisors build and manage factor portfolios. The concept of mapping a strategy to a market environment is known and widely utilized.

Making a sweeping statement that researchers can predict returns has been a bit more of a stretch, and here the authors break new ground.

To accomplish this previously unattainable feat, the study categorizes all strategies into one group and measures a set of 13 “predictive variables” that are tested through three primary market environments.

The 13 predictive variables included common trend following pattern market environment benchmarks such as a moving average cross, in this case a long time horizon 200 day and 20 day simple moving average cross (the industry standard for benchmarking the market environment of price persistence is often the Societie Generale 120 / 20 day moving average formula). It is worth noting that a trend following is not considered forward-looking. The strategy is seldom attempting to “predict” as much as it is understanding win percentage relative to win size. When studying individual trades, trend following can have a win percentage below 50%, but it is on win size of the winners where the strategy has historically done better on a relative basis. That said, the moving average cross methodology has been used in a variety of studies to determine if the market environment was positive or negative for stocks.

The moving average cross was one of several factors such as price-earnings ratios and dividend yield statistics relative to volatility and interest rate benchmarks that determined market environment.

The report nonetheless took these predictive variables and modeled them through three market environments: a “bear market” from January 2007 until February 2009; a “recovery” market from March 2009 to August 2012 that spanned the European debt crisis through what they called the “Draghi-Put” environment; to the third, September 2012 until June 2015, which they considered a bull market period.

“We try to predict one-months ahead returns using a set of different predictive variables / models,” Tille said. They in part drew their conclusions by looking at a fund of fund portfolios that made hypothetical investments into the 10% of funds with the highest return forecasts and compare the performance of the predictors with using historical fund returns only.

While they make claims on the ability to have developed a “predictive” model, the caveats point to different conclusions. “Overall, we cannot conclude that a more robust estimation approach would generally be better for investors,” the report stated in its conclusion when discussing the ability to “capture distributional information as non-linear relationships between predictors and fund returns.”

Tille explains this assessment of the formula is in relation to the question whether to use OLS or quantile regression to estimate the predictive models. “There was also no clear advantage of using either combination forecasts, a highly parameterized kitchen sink model or regularized lasso and ridge regression models,” he said, trying to clarify that using one of these approaches (combination, lasso/ridge or kitchen sink) are more sensible than relying on single predictor models.”

Perhaps the hot topic of market environment correlations is the more interesting topic, not “predicting” anything

While their results and methodologies might require deeper examination, the report touches on the hot topic of modeling various strategies to a market environment. Multiple quantitative research sources have told me this work is going on behind the scenes at several large equity-based firms in New York and Chicago, although this has not yet been publicly confirmed.

In the managed futures space and algorithmic portfolio building world, the concept of recognizing market environment and mapping strategies is not new.

Typically, alternative strategies are first broken down into systematic and discretionary strategies because they model differently. A systematic strategy, based on a formula, follows a Boolean if-then logic and is seeking to exploit repeatable market events. A discretionary or fundamental strategy cannot be as easily predicted because, unlike a formula, there is no certain outcome given similar inputs.

A systematic strategy is based on a formula and follows a Boolean if-then logic and is seeking to exploit repeatable market variables. A discretionary or fundamental strategy cannot be as easily modeled because, unlike a formula, with humans, there is no certain outcome even when given similar inputs. Systematic managers can claim with a degree of certainty that if a market environment of price persistence is a feature of markets over a specific time frame, then the strategy is expected to model positively to various degrees. When price persistence is not a feature of markets, the trend strategies perform poorly. The same holds true of a market environment of Volatility and Mean Divergence/Convergence.

The differences between fundamental/discretionary strategies involving human logic that lead to very different modeling outcomes. The report, however, did not appear separate human discretionary strategies from systematic, formula-driven strategies in the study.

What the paper does show is that they claim to model generic strategies as one monolith through a series of market environments, with an emphasis on Long / Short and alternative fixed income strategies.

In some managed futures CTA categorization methods, the Long / Short strategy is placed in the Relative Value section and has a specific market environment to which it correlates. Long / Short relative value players are known to benchmark the beta market environment of price relationships that divert from their statistical mean and then revert back. Sometimes trend following strategies are utilized inside a relative value strategy “tracking the trend of a price dislocation and its relative force“ and adjusting a strategy based on the force of trend in the beta market benchmark.

This type of research is typically done without unnecessary mathematical obfuscation and has often avoided the moniker of predicting returns.  Measuring a price divergence is a transparent and objective methodology that can be explained in a few sentences.

Certain practitioners in the algorithmic portfolio building space have concluded that a strategy can be modeled to a market environment to various degrees – trend following can obviously be modeled to price persistence over different time horizons, for instance — but predicting returns requires understanding market environment comes next. Here is the challenge with the “predicting” component and where difficulty meets reality and lower win percentages are visible.

In making bold claims regarding predicting returns, the researchers really point to the current hot topic of how a macro market environment impacts alternative strategies.

Can they do it? Read the full report here and decide for yourself.

Insight / Analysis / Opinion:

There are examples of how a strategy can model through different market environments that can be done without overly complex formulas. On a most basic level, a trend following system performance can be approximated based on the market environment of price persistence, the technical price pattern and length and consistency of the trend. Different types of technical pricing patterns can produce expected returns during different types of market environments. But predicting which market environment will occur next — the key to “predicting” returns — is a difficult if not impossible task on a consistent basis.