Stress Testing CTA Portfolios – Impact of Volatility Adjusting Positions

In light of recent market performance, and the corresponding effect on changes in volatility on CTA returns we thought it important to give our views on the topic. Late last year, we were asked by a prospective client to see how one of our trend following models performed over several different stress environments. We highlight one particular stress that was given- a 20% stock market drop over 3 months, with 40% of move in month 1, 35% in month 2, and the last 25% of the move in month 3. A relatively straightforward exercise, but to really understand the nuances of different CTAs relative to our approach, you must look past just the change in level, but consider the potential price paths and volatility over that stress period. The difference boils down to whether one is viewing CTAs as a standalone investment, or as a piece of a larger portfolio, and the role of volatility targeting in position sizing.

Chart 1 illustrates a continuous futures price series for the S&P 500 in the 100 days ending September 30, 2017, then 4 different hypothetical price paths to simulate the stress period over the next 3 months. The initial case, a “Linear” (blue line) price move, where the price moves the same amount each day. A “OneDayDrop” (orange line) case where the price stays flat till the end of each month and makes the entire move on the last day. A “WhipSaw” (grey line) case where the price moves up and down each day finishing the month with a down move. Last, an “ExpandedVol” (yellow line) price time series case with elevated market volatility defined set at 32%. They all get to the same end point.

stresstest1

Chart 1. S&P 500 Stress Period Price Paths. Past performance is not indicative of future results.

We then ran our trend following model, the “Base Model,” over each of the price paths and charted the performance in Chart 2. The Base Model follows a simple long/short trend following signal and the positioning is based entirely on the strength of that signal. Explicitly, the position does not rely on underlying market volatility at all. The performance begins on the first day of the stress period and runs for three months. As you would expect, as prices begin to fall, the Base Model loses money in each case, but as positions move from long to short, each model begins to see the corresponding gains. In short, there isn’t a huge amount of difference in the performance of each price path, they all end up around the same place. This is how we approach trend following at Mount Lucas.

stresstest2

Chart 2. Trend Following Performance – Base Model. Past performance is not indicative of future results.

Now, as a different exercise, let’s run the same model over the same price paths, and add a simple position function that is determined by volatility at the market level- taking larger positions in low volatility environments, and smaller positions in high volatility environments, the “Vol Model”. This is a very common method used among other CTAs. The picture is very different, and now the different price paths really matter. The “Linear” decline will take volatility down to almost zero, in this case positions will grow, and the model will likely be more profitable in this perfect world. The model will see risk as almost nonexistent (in reality there is usually a floor of volatility or exposure cap that will prevent positions growing exponentially – but the point still holds – positions will increase meaningfully before that floor/cap is hit). The realized volatility levels of the paths are 1% for the “Linear” move, 23% for the “OneDayDrop”, 112% for the “Whipsaw” move, and then 32% for “ExpandedVol”, the more realistic scenario, which we will concentrate on here. Chart 3 compares the position sizes of the Base Model and the Vol Model during the stress period in the case of the “ExpandedVol” price path.

stresstest3

Chart 3. S&P 500 Position Size – Base Model vs Vol Model: ExpandedVol Price Path. Past Performance is not indicative of future results.

The lines in Chart 3 represent the position sizing of two different approaches run on the same underlying trend following signal model. The Base Model does not adjust the position for the underlying market volatility, it gets short and is driven by just the price action, and then stays there. The Vol Model scales the positions by the underlying market volatility. As it moves up from the starting period volatility of 8% or so (this would change with parameters, but for illustrative purposes we use 20 day realized volatility) to 32% volatility, the short positions get reduced by around half.

Chart 4 shows how the position sizing of the Base Model (blue line) versus the Vol Model (orange line) affects performance during the stress period along the “ExpandedVol” price path (grey line). Both models get hurt by the early drop in prices, as both start the stress with similar long positions. As the market keeps declining, both models start to take short positions, but the Vol Model takes smaller positions as volatility has also increased. While the Base Model maintains position size and makes back the early losses and finishes with gains, the Vol Model is under exposed and finishes down over the period.

stresstest4

Chart 4. Trend Following Performance – Base Model vs. Vol Model: ExpandedVol Price Path. Past performance is not indicative of future results.

In our view, a decision to choose between the two approaches is really one to be made in a portfolio context. Generally speaking, trend following investors allocate a portion of their portfolio to the strategy as a diversifying asset to balance equity and credit investments. If one is making a standalone portfolio allocation just to trend following, a volatility adjusting strategy will likely produce more steady returns, with fewer and smaller positive tails. However, in a world where the trend following allocation is part of a larger portfolio with equity and credit, the better option is to maximize the positive tail and accept possibly lumpier returns, as the value of the positive tail when you need it is greater. In terms of matching return distributions, where equity and credit have negative tails, wouldn’t you want your balancing asset to maximize its positive tails? Similarly, if investing in a portfolio of trend following strategies where most use some method of volatility adjustment, adding another that does the same to us makes less sense than adding one that is different.

Another way to conceptualize this, think of trend following investing as buying a call on large directional market moves, which you will see in a crisis. Volatility adjusting positions downwards due to rising volatility reduces the crisis call delta, at exactly the time the rest of an equity and credit portfolio is getting hurt. You end up with a call spread instead of an outright call – truncated crisis upside. It makes little sense to hold a diversifying asset for a crisis period, only to have it reduce positions at the first sign of panic.

The scenarios discussed contain hypothetical performance results and are for discussion purposes only. Hypothetical performance results have many inherent limitations. No representation is being made that any account will or is likely to achieve profits or losses similar to those shown. There are frequently substantial differences between hypothetical performance results and the actual results subsequently achieved by any particular trading program.  Additionally, Mount Lucas employs a number of different strategies each with their own investment objectives and risk profiles.  Any reference to a strategy or strategies mentioned above may or may not be indicative of all of Mount Lucas’ products.