A Dead Simple 2-Asset Portfolio that Crushes the S&P500 (Part 4)

/

For newsletter subscribers this post is best viewed directly on my blog.

Recap

In Part 3 of the series we reviewed the relationship between returns and correlation of the 2-asset portfolio UPRO and TMF. The basic equal weight strategy was very compelling in terms of total return and CAGR. However, the strategy is susceptible to large drawdowns, especially in situations where US equities and long term bonds are out favor, for example in the 2015 and 2018 periods. We also went over some prototype strategies while introducing two indicators in an attempt to improve the risk-adjusted returns. There were several strategies that showed enough promise to be worth investigating using a more professional backtesting engine.

Introduction

In this blog post we will review the simulated performances of a few UPRO/TMF strategy implementations using the Quantconnect platform. If you’re not familiar with the platform, it is an algorithmic trading platform that provides backtesting and live trading across of variety of asset classes including: equities, futures, forex, options, and cryptocurrencies. I like using the platform because of the access to a large number of asset classes, the development team is responsive and you can code strategies in Python (even though the underlying platform is built in C). The strategies’ performances are evaluated using pyfolio and ffn. Note that in some cases their calculations are slightly different.

Outline

  • Strategy Description

  • Benchmark Results

  • Basic Strategy Results

  • Basic Strategy + Risk Management Results

  • Secret Sauce Results

  • Conclusion

  • Future Work

  • Disclaimers

  • References/Resources

Strategy Description

Here is an outline of the basic parameters of the various strategies. The strategies implemented in Quantconnect are variants of the equal weight, equal risk contribution weighting schemes. Some of the strategies use the Momersion indicator, some use the MMI Indicator. The parameters:

  • Covers the period from 2011-01-01 to 2019-04-30.

  • Starting capital is 100,000.

  • Lookback is 252 days (not-optimized).

  • Any indicator windows are 120 days for the long period and 50 days for the shorter (not-optimized).

  • Daily Rebalance with a tolerance band of +/- 2.5%, (pretty high turnover).

  • Rebalance is scheduled daily at 10 minutes after market open for UPRO (not-optimized).

  • Uses Interactive Brokers commission model to estimate transaction costs.

Benchmark Results

ALL RETURNS HYPOTHETICAL AND SIMULATED. PAST PERFORMANCE DOES NOT NECESSARILY PREDICT FUTURE PERFORMANCE.

ALL RETURNS HYPOTHETICAL AND SIMULATED. PAST PERFORMANCE DOES NOT NECESSARILY PREDICT FUTURE PERFORMANCE.

ALL RETURNS HYPOTHETICAL AND SIMULATED. PAST PERFORMANCE DOES NOT NECESSARILY PREDICT FUTURE PERFORMANCE.

ALL RETURNS HYPOTHETICAL AND SIMULATED. PAST PERFORMANCE DOES NOT NECESSARILY PREDICT FUTURE PERFORMANCE.

The benchmark strategies’ performances are pretty strong on a risk-adjusted basis, especially for such a simple implementation. However as previously highlighted, the drawdowns are significant with max and average of the 5 worst drawdowns exceeding our -15% target. Also the average of the 5 worst drawdown durations are much greater than our target of 120 days (~ 4 months), extending to 200 days or more.

Basic Strategy Results

ALL RETURNS HYPOTHETICAL AND SIMULATED. PAST PERFORMANCE DOES NOT NECESSARILY PREDICT FUTURE PERFORMANCE.

ALL RETURNS HYPOTHETICAL AND SIMULATED. PAST PERFORMANCE DOES NOT NECESSARILY PREDICT FUTURE PERFORMANCE.

ALL RETURNS HYPOTHETICAL AND SIMULATED. PAST PERFORMANCE DOES NOT NECESSARILY PREDICT FUTURE PERFORMANCE.

ALL RETURNS HYPOTHETICAL AND SIMULATED. PAST PERFORMANCE DOES NOT NECESSARILY PREDICT FUTURE PERFORMANCE.

SELECTED TOP PERFORMING BASIC STRATEGIES

ALL RETURNS HYPOTHETICAL AND SIMULATED. PAST PERFORMANCE DOES NOT NECESSARILY PREDICT FUTURE PERFORMANCE.

Of the basic strategies, strategy number 9 was the only strategy to avoid a losing year while having a significant amount of trading in each year. From a risk-adjusted perspective it had one of the lowest calmar ratios of the top basic strategies and it’s average 5 worst drawdown durations was longer than 180 days. Strategies 19 and 14 had great risk-adjusted metrics however there were periods of significant inactivity.

Basic Strategy + Risk Management Results

ALL RETURNS HYPOTHETICAL AND SIMULATED. PAST PERFORMANCE DOES NOT NECESSARILY PREDICT FUTURE PERFORMANCE.

ALL RETURNS HYPOTHETICAL AND SIMULATED. PAST PERFORMANCE DOES NOT NECESSARILY PREDICT FUTURE PERFORMANCE.

ALL RETURNS HYPOTHETICAL AND SIMULATED. PAST PERFORMANCE DOES NOT NECESSARILY PREDICT FUTURE PERFORMANCE.

SELECTED TOP PERFORMING BASIC + RISK MANAGEMENT STRATEGIES

Adding a risk management component to the strategies helped reduce the best performing versions max drawdown to under 20%. Calmar is still greater than 1. Also the average of the 5 worst drawdown durations are very close to our target of less than 120 days, and the average of the 5 worst drawdowns is less than 15%.

Secret Sauce Strategies

ALL RETURNS HYPOTHETICAL AND SIMULATED. PAST PERFORMANCE DOES NOT NECESSARILY PREDICT FUTURE PERFORMANCE.

ALL RETURNS HYPOTHETICAL AND SIMULATED. PAST PERFORMANCE DOES NOT NECESSARILY PREDICT FUTURE PERFORMANCE.

ALL RETURNS HYPOTHETICAL AND SIMULATED. PAST PERFORMANCE DOES NOT NECESSARILY PREDICT FUTURE PERFORMANCE.

ALL RETURNS HYPOTHETICAL AND SIMULATED. PAST PERFORMANCE DOES NOT NECESSARILY PREDICT FUTURE PERFORMANCE.

SELECTED TOP PERFORMING SECRET SAUCE STRATEGIES

ALL RETURNS HYPOTHETICAL AND SIMULATED. PAST PERFORMANCE DOES NOT NECESSARILY PREDICT FUTURE PERFORMANCE.

Conclusions

The performance of the simple UPRO/TMF strategy across various implementations is extremely compelling. The most basic implementation does crush the performance of the SPY ETF on a risk-adjusted basis. Some causes for concern are that the basic strategy struggled in 2015 and has been choppy since 2017-2018.

Adding a simple risk management component helped dampen the drawdown magnitude and duration of a couple of the highlighted variations.

Finally, the secret sauce versions, which build on the basic strategy with risk management, had the most consistent performance across the backtest period while also limiting max drawdowns to less than 10% and maintaining calmar ratios greater than 1. Also note that all 3 top performers from this group had a positive skew at the monthly level. Strategies 3 and 18 also maintained an average of 5 worst drawdown durations of less than 120 days.

Depending on one’s risk tolerance and capital allocation there’s a variation of this strategy that is likely to be beneficial, either as a standalone or complement to a portfolio of strategies. What makes this strategy particularly interesting is that it stands on an economically understandable foundation. In periods of equity dominance, large caps generally do well, with the leverage of UPRO outpacing the declines or choppiness of TMF. During periods of equity uncertainty, generally, long term bonds are bid and the TMF performance is likely to offset or dampen the effects of a declining UPRO. There are also many situations when both equities and long term bonds are bid, in which case any variation of this strategy will outperform as both assets will trend higher with the 3x daily leverage magnifying the portfolio returns. The downside, as previously discussed, are those rare scenarios when US equity and long term bonds are both out of favor, in which case almost all variations of this strategy will suffer. The secret sauce variations are likely to handle those situations much better and in some cases still maintain positive absolute performance. Worst case scenarios for all versions are linked to a US economic collapse where large cap equities and US treasuries are both sold in size over an extended period of time.

Future Work

I’m open to collaborating with interested parties to implement this strategy live. If you would like to personally utilize any of the implementations displayed here contact me via email at bcr@blackarbs.com or fill out the form at the bottom of the services page.

Disclaimers

Results are hypothetical and simulated and past performance does not necessarily predict future performance. Efforts have been made to ensure realistic simulations by including simulated transaction costs and slippage, however live performance may differ from simulated performance due to many factors that include, but are not limited to: rebalance frequency, more or less slippage, more or less commissions, margin restrictions, capital allocation, and future market environments.

Efforts have been made to reduce data snooping and parameter optimization. Parameters were not optimized, however there was an alternate variation tested where the long indicator window was set at 250 days. In most simulations the change in window from 250 to 120 days had almost no impact, while the performance of a small number of variations did improve.

The risk management component does have 3 parameters that are adjustable based on risk tolerance, however those parameters were not optimized for maximum performance.

References/Resources