A Global Lookout To Quantify And Mitigate Risk For Crash Protection

|
Includes: BIL, EEM, EFA, GLD, GSG, HYG, IEF, IYR, LQD, QQQ, SHY, SPY, UUP
by: TrendXplorer
Summary

Numerous quantitative strategies exploit momentum: the well-studied tendency for asset prices to keep moving in the same direction.

However, even momentum approaches show limitations by exhibiting substantial risk in the form of severe drawdowns, losing calendar years, negative 12-month rolling returns, and increased volatility.

To reduce risks yet preserve returns, our "Protective Asset Allocation" (PAA) approach combines dual momentum with vigorous crash protection based on multi-market breadth.

Using this algorithmic estimator PAA mitigates risk considerably and also produces superior risk-adjusted, positive returns with high consistency.

We highlight how PAA works, describe backtests covering 45 years, and provide references where readers can learn more about the particularities of PAA.

Introduction

Obtaining consistent positive returns is possible by combining a dual momentum based tactical asset allocation model with a vigorous crash protection algorithm. Our crash protection is based on a multi-market breadth indicator for quantifying risk in global markets. This innovative "protective momentum" approach is demonstrated in a recent SSRN-paper by Wouter Keller and me titled "Protective Asset Allocation". In our PAA-paper we used a simple moving average based dual momentum strategy for timing as well as selection. What's more, our multi-market breadth crash protection approach can easily be combined with other quantitative methods for portfolio allocation.

Dual momentum

The dual momentum approach has been popularized in publications such as those by Antonacci, Faber, and others. They combined two characteristics of momentum: absolute and relative momentum. Momentum is based on the assumption of price persistence: rising prices will continue to rise and falling prices will get even lower. Momentum is often measured by past returns over, for example, the prior 12 months. Assets below an absolute momentum threshold, like a simple moving average (SMA) price filter, are replaced for a low-risk safety asset. Relative momentum compares momentum among assets for selection of the top assets with the highest momentum readings.

One of the many dual momentum strategies is Mebane Faber's famous Global Tactical Asset Allocation model (GTAA, updated in 2013): allocate capital in (equal) portions to the assets in the top momentum selection of a universe that are above their long-term SMA, invest the remainder in a safe haven short duration treasury ETF like SHY, and apply monthly portfolio reforms. For simplicity, our take on GTAA deploys the ratio of current price to its 1-year SMA minus one for both absolute and relative momentum.

So this momentum metric governs both timing: ratio above zero, and selection: the highest ratios.

The backtests in this article are for a globally diversified universe with ten "risk-on" ETFs: SPY, QQQ, EFA, EEM, GLD, GSG, IYR, UUP, HYG and LQD with a separate safety or "risk-off" asset for crash protection. None of these ETFs have a price history dating back to 1970. To obtain these synthetic ETF-proxies, we constructed monthly total return price series based on indices net of costs (see PAA-paper Appendix for details).The backtests cover the period December 1970 until March 2016.

The following equity chart visualizes in semi-log scale GTAA's performance over 45 years for a top selection of three out of ten assets with SHY as a safety asset. The table below the chart holds some key performance metrics.

NB! All results in this article are derived from synthetic monthly total return data constructed by us based on indices net of costs. In the backtests here I disregard trading costs, slippage and taxes. Results are therefore purely hypothetical.

While GTAA's raw compounded average growth rate R is nothing less than impressive, its risk profile is rather disquieting with an annualized volatility (V) of 14% and a maximum drawdown (D) of 30%, despite its Sharpe ratio (SR: R/V) of 1.13. MAR (R/D) is 0.53. With time spent in drawdown at 60%, the strategy was 324 months underwater (out of 540 months), while the longest drawdown period was more than 4 years on stretch (50 months). The number of negative years is 7 with -13.5% being the worst year.

With only 88% respectively 94% of 1-year rolling returns above 0% respectively above -5% GTAA scored (way) below the goals we aimed for in our paper: 95% above 0% and simultaneously 99% above -5%. Therefore GTAA's performance cannot be qualified as consistently positive. Rolling returns are important for evaluating strategy robustness, because, contrary to trailing returns, rolling returns test the consistency of a strategy's performance over time.

GTAA's downside risk profile is illustrated by the following chart painting its 1-year rolling returns. Notice the severe drops below the -5% mark on several occasions.

The root of the discussed limitations lie in the way downside risk is handled. In theory GTAA does have a form of crash protection on alert: when one or more assets in the top selection are below their long-term moving average it is replaced by the safety asset. However, in practice protection is virtually non-existent, at least for a concentrated top selection like the one at hand with three out of ten assets. This is demonstrated by the following diagram of GTAA's capital fraction (%) to the safety asset.

The safety asset SHY is deployed during only 4% of the backtested months, resulting in an average capital allocation of below 2%. The reason is that the smaller the top selection is, the less likely deployment of the safety asset becomes, since only the top selection is checked on absolute momentum.

Protective momentum

To counter the above mentioned limitations of the traditional dual momentum approach, we designed "protective momentum": an adaptive algorithmic estimation of risk based on multi-market breadth applied to global asset-classes. We use this breadth indicator as lookout for imminent market crashes by quantifying and mitigating risk. Therefore the full asset universe is scrutinized for signs of weakness instead of only the top selection. With various protection levels to choose from a tailored risk/return profile can be reached. The following table demonstrates the degree of capital preservation for three protection levels (low, medium and high) applied to a universe with ten "risky" assets, irrespective of the top selection size.

The difference between the traditional top only approach compared to PAA's full universe approach for crash protection becomes clear with the following example. Consider a universe with ten "risky" assets from which each month the three assets with highest (relative) momentum are selected. Suppose in this universe five assets register positive momentum and the other five assets have a momentum reading below zero. In such a case the traditional take would see no cause to deploy the safety asset, since all assets within the top selection have positive momentum (3 < 5), while protective momentum would assign 50% - 100% of capital to the safety asset depending on the protection factor setting (see the above table's middle [5, 5] column).

To crystallize the effect of PAA's vigorous crash protection algorithm, the below diagram is instructive. The diagram shows the capital allocation to the safety asset for PAA with the highest protection setting on guard. The backtested universe and top selection are equal to the ones used for GTAA's diagram above.

For the scenario with the protection level set to "high" the safety asset is deployed nearly constantly: 97% of time compared to only 4% for GTAA. Furthermore the average capital fraction (%) to the safety asset is 57% instead of 2% for GTAA.

Notice also the waxing and waning of the safety asset's capital allocation during equity bear and bull markets, which is the effect of the multi-market breadth algorithm by quantifying risk in the ten global markets present in our testbed.

Like every form of solid protection, there is a premium to be "paid" as backtesting various PAA scenarios will show.

Backtesting PAA scenarios

The backtests of PAA are performed with the earlier introduced diversified "risky" universe representing ten global markets. Just like GTAA each month the three out of ten assets with the highest momentum ratios are selected for capital allocation next to the safety asset's allocation. On the chart below PAA is shown with low (PAA0), medium (PAA1) and high protection (PAA2) enabled and compared to GTAA, each with a top selection of three out of ten assets. For all scenarios, SHY features as safety asset. The equity curves are depicted in "wet paint fashion" to emphasize the magnitude and duration of drawdowns. The chart also is on semi-log scaling. (The numeral identifiers "0", "1", and "2" are references to our breadth formula. For details please see our paper.)

Compared to GTAA's performance the "raw" compounded return is 1.3%-2.8% lower for the various PAA scenarios. However, all risk measures deteriorate monotonically when going from PAA2 (high protection) over PAA1 (medium protection) and PAA0 (low protection) to GTAA's (top only protection). Notice only PAA2 satisfies our positive return requirement: the above 0% and -5% readings for its 1-year rolling returns are 96.80% and 99.81% respectively. The drop in return can be considered as the "insurance premium" for PAA2's vigorous capital preservation as crash protection.

To demonstrate another feature of PAA, with the protection level set to "high" the model is again backtested three times, but now each time with a different safety asset: SHY, IEF and the SHY/IEF combination. For the last scenario instead of using a fixed safety asset an alternating safety asset is deployed: each month the top pick out of SHY and IEF is selected for capital preservation. The safety asset selection is based on our earlier specified momentum ratio, irrespective of the sign. The resulting three equity curves are shown on the following chart in semi-log scale with again some key performance metrics in the table below the chart (coloring does not match previous chart).

The chart visualizes how in particular the PAA2 scenarios with IEF and SHY/IEF as safety asset preserved return quite well. The chart also shows that the scenario with the alternating safety asset SHY/IEF proved to outperform the scenarios with a fixed safety asset for especially the first decade of the backtest: years with overall rising treasury rates. This demonstrates that protective momentum is a viable approach in a rising rate environment too. To strengthen yield neutrality even more, the safety asset combination could be expanded to, for example, BIL/SHY/IEF.

The preferred strategy is PAA2 with SHY/IEF deployed as safety asset, since this scenario delivered positive 1-year rolling returns with the highest consistency: the above 0% and -5% readings for its 1-year rolling returns are 97.56% and 99.62% respectively. These 1-year rolling returns are charted below followed by the isolated equity graph of this scenario. Only in 1987 (Black Monday) and 1994 (Bond Crash) we see the 1-year rolling returns drop markedly below 0% with only one year (1994) where the 1-year returns just dipped below the -5% mark.

The equity chart above visualizes in semi-log scale PAA's performance over 45 years for a top selection of three assets out of ten assets with SHY/IEF alternating as safety asset.

Google Spreadsheet

A quantitative strategy like PAA can be coded for a platform like AmiBroker as used for the charts in this article, but setting it up in Excel or in Google Spreadsheets is by all means doable. This is demonstrated by the "live" updating Google Spreadsheet version of PAA as described in this article: PAA for selecting three out of ten "risky" assets with the protection level set to "high" and SHY/IEF alternating as safety asset. The screenshot below is taken after April's close with the signals and position sizes for May: 100% IEF.

Since by design PAA is purely mechanical, there is no need second guessing market conditions nor predicting trends. Therefore the result may appeal to many retail investors: close to zero maintenance/attention, easy execution, and total objectivity.

Conclusion

The scope of this article was to present the highlights of PAA. PAA is a new tactical approach to harvesting momentum that incorporates innovative, effective crash protection strategies over and above those typically used in dual momentum strategies. We demonstrated PAA's key feature: a multi-market breadth algorithm to quantify and mitigate risk in global markets. Backtested performance over the last 45 years demonstrate that PAA's risk mitigation has been effective, without compromising return on investment. We also showed PAA's ability to obtain positive 1-year rolling returns consistently. Lastly we illustrated PAA's mechanical design through a Google Spreadsheet example.

Anyone looking for comprehensive knowledge about PAA or (dual) momentum strategies in general, is referred to the following sources as suggestions for further reading:

Disclosure: I am/we are long IEF. I wrote this article myself, and it expresses my own opinions. I am not receiving compensation for it (other than from Seeking Alpha). I have no business relationship with any company whose stock is mentioned in this article.