ETF Complete Strategy Insights: Slicing Up the ETF Models (Part 2)

James Kimball | August 25, 2019

The ETF Complete model closed the week up +0.6% compared to the SPY which closed down -1.4%.

After a quiet start to the week, Friday saw a re-injection of volatility into the markets with most indexes closing back near the prior week's low. Trade and tariff woes were the primary drivers of Friday's selloff.

Stay tuned for the daily updates and log into the website to see holdings and additional performance data.

image

This Week's Strategy Lesson: Slicing Up the ETF Models (Part 2)

image

This week we are going to wrap up this brief series on probability and expected value. As we covered last week, knowing and setting the right expectations for a trading strategy is very important, both in terms of the initial decision of whether to trade the model and how much capital to deploy, to helping you with stay committed to consistently executing the model.

The ETF models are more than just a collection of unrelated trades. There is a consistent selection methodology that limits the universe to relevant ETFs that have strong trend characteristics. All of the trades are managed using the same set of rules. And the position sizing and rebalancing rules further set them apart.

Of course, it’s impossible to predict the outcome of any individual trade, and the performance characteristics of the whole can change overtime with new data, however, the list of ETF trades is over 500 and includes trades during a variety of different market cycles and events.

image

The basic trading stats in the table above are for two main ETF model variants. While each model has different values, they follow similar patterns. Both have around a 50%-win rate. Both have about a 2-1 ratio in the size between their average wining trades and average losing trades. And both have a profit factor around 2.

One of the main differences between the models is the scale of the average winning and losing trades. The Complete model tends to have lower volatility than the Sector model. The main reasons for this is that the Sector Moderate model is both more concentrated (3 positions instead of 9) and tends to employ more leverage.

image

The chart above returns to the probability distribution we looked at last week, but now with the Sector models and the SPY added in. The SPY uses monthly returns instead of trades. It is far from a perfect equivalent but should work fine for our purposes here.

To review, the chart above takes all the trades from each of the models (or monthly returns for the SPY), sorts them by return, separates the trades into 10 equal buckets, and finds the average return for all the trades in each bucket.

One way to look at the chart above is to imagine you were playing a game where you were randomly pulling out marbles from a jar blindfolded. In the jar, there are 10 marbles, each with a different number printed on it corresponding to the 10 average trade return buckets.

Essentially, this marble picking process simulates the potential returns we might get from any one trade chosen randomly. While we would like to get one of the better outcomes, for any one trade, the results could fall anywhere on this spectrum. As our marble or trade sample size increases, we would expect the results to start to look like the chart above.

To convince a normal, rational person to play this game, the game needs to have a “positive expectancy.” At the end of the day, the player needs to know or believe they will be up, whether by winning more often than losing or winning more when you win than lose.

The probability distribution in the chart fits this bill. While it is close to 50/50 on winners versus losers, the magnitude of the difference between the average size of the winning versus losing trades is large at around 2-1. There is some variance, as the Sector model has larger winners and larger losers than the Complete, but the ratio between the size of the average win and the size of the average loss stays about the same.

Understanding the probabilities and outcomes of a model can help us set the right expectations and stick to the rules, both when the model is hitting on all cylinders and when we experience the inevitable pullbacks, knowing that the long-term odds are in our favor.