Adaptive Vs. Static Asset Allocation: Addressing Potential Portfolio Losses

by: John M. Balder

Summary

Investment parameters (returns, risks and correlations) are not stable over time.

Static "diversified" approaches to asset allocation have generated significant losses since 2000.

An adaptive approach can be employed to mitigate future losses.

Given risks of potential losses, we will argue that right now may well turn out to be precisely the wrong time to invest in a passive "buy-and-hold" investment portfolio. Even John Bogle, founder of Vanguard and a strong advocate of passive investing, acknowledges that investors are likely to experience one or two periods of significant losses (25% to 50%) over the next decade. We agree -- our response is to create an adaptable investment process, with a focus on mitigating downside risk. Why?

Let's take a look at how buy-and-hold management has fared since the turn of the millennium (see chart below). This chart isolates the downside risks associated with a buy-and-hold investment in the S&P 500 and an equal-weighted portfolio (equities, bonds, commodities and REITs) since January 2000.

An allocation to the S&P 500 would have generated a loss of 62% between January 2000 and March 2009. An allocation to a more diversified equal-weighted portfolio would have declined by 38% during the 2008 crisis, having declined earlier in the decade by 14%. A loss of 38% would require a gain of 61% just to get back to even. Buy-and-hold investing exposes investors to potentially significant drawdowns. It is interesting to note that even a "diversified" (equal-weighted), buy-and-hold portfolio does not protect investors from major drawdowns. Think about that the next time your investment manager tells you that they are "diversified."

Back in the 1950s when Harry Markowitz introduced modern portfolio theory, he cautioned against using long-term estimates of excess return, volatility and correlations (though this concern was drowned out by the acclaim that greeted the new framework) in allocating portfolios. In today's marketplace, the drawdowns described in the chart above (especially for the equal-weighted portfolio) speak to the foibles of using long-term estimates in constructing portfolios (unless you have a horizon of between 50 and 100 years AND have the ability to stick with your allocation). And the risk or likelihood of significant losses cannot be ruled out today, given the continued growth in various imbalances (excessive debt to income ratios and overvalued asset prices) as well as a potential end to aggressive central bank policies.

We have studied the instability of the three parameters from 1900 until today. Below, we focus on their behavior from the early 1980s, using the S&P 500 as a case study. We will explore whether the use of long-term estimates of returns, volatility and correlations are reliable.

Portfolio Returns: S&P 500

We begin by assuming that an "average" investment horizon is between three and five years. From 1980 to 2016, three-year rolling returns for the S&P 500 ranged from -16% per annum to 30% per annum (see chart below). If, for example, we had chosen to invest in the S&P 500 in March 2006, three years later we would have endured an annualized loss of more than 16%. Our investment would have declined by a total of 41%. With a horizon of 5 years, in March 2009 we would have lost -8% per year, resulting in an overall loss of about 31% of the investment. And given behavioral drivers, including asymmetric loss tolerance in financial markets, it is likely that a buy-and-hold investor would liquidate, thus missing out on the recovery that began in 2009.

Volatility: S&P 500

Next, we assess rolling three- and five-year volatility using the S&P 500 (similar data for other classes are available from the author). Rolling three-year volatility ranged from 5% to 20% over the 1980 to 2016 period, while rolling five-year volatility ranged from 6% to 17%. Here again, volatility measures are subject to significant fluctuations. Investing based on a historical volatility estimate also is not reliable.

Correlations: S&P 500 and other Asset Classes

Rolling correlations between the S&P 500 and other asset classes also fluctuated throughout the 1980 to 2016 period. The table below examines the range (high/low) of correlations between the S&P 500 and other asset classes. (Correlation coefficients between two asset classes can range from -1.0 to +1.0.) The range of correlations below indicate that once again a point estimate is not reliable.

Why do correlations matter? In the Appendix, we calculate a hypothetical portfolio that is evenly divided between equities and bonds. The Sharpe Ratio (returns in excess of cash divided by volatility) for this portfolio will range between 0.36 and 1.80, depending on the correlations. The difference in terms of returns per unit of risk is quite substantial. So correlations matter a great deal and to the extent that they fluctuate, we should develop a method for capturing that fact.

Exploring Unstable Parameters

We have demonstrated above that "buy-and-hold" parameter estimates provide unreliable forecasts. Below, we take a further look at two interactions between the parameters.

The sharp decline in equity valuations is often accompanied by a spike in volatility (see chart below).

The sharp decline in equity valuations in 2008 was closely correlated with the spike in correlations ("correlations head to one in a crisis").

In summary, declines in excess returns are often accompanied by spikes in correlations and volatility. This is another important connection that should be taken into account in developing an investment framework.

ADAPTIVE FRAMEWORK

We have illustrated above that parameters are unstable and fluctuate over time. Creating a buy-and-hold investment portfolio based on long-term average parameter estimates exposes portfolios to potentially significant downside risks. Some advisors nonetheless recommend that clients stick with the strategy (based on their belief that these drawdowns represent a "black swan" and are impossible to predict -- having endured the downside, why not enjoy the upside).

We think there is a better way: namely, creation of portfolios that adapt to change (based on an August 2004 article entitled the "Adaptive Markets Hypothesis," by Andrew Lo). This can be achieved using one or more of the following approaches: A regime-based approach that adapts to changes to macro-financial conditions; a momentum (returns-based) approach ("today's value is the best predictor of tomorrow's"); a risk-parity approach (equalizing risk to provide greater portfolio diversification -- generally resulting in lower equity exposure and enhanced diversification).

Appendix: Why Correlations Matter?

In this simple example, we assume the following attributes: only the correlation varies.

The average portfolio return = 0 .5 *3.00% + 0.5*6.00% = 4.5%

The Sharpe Ratio (return in excess of cash divided by the portfolio standard deviation) is in the following table:

Interpretation:

Maximum Diversification: Perfect Negative Correlation

when the correlation is perfectly negative (-1.00), the portfolio risk = 2.5% and the Sharpe Ratio = 1.80, meaning 1.8% in excess return for every 1.0% in risk.

Some Diversification: No Correlation

When the two assets are not at all correlated (0.00), the portfolio risk = 9.0% (less than the average of the two asset classes) and the Sharpe Ratio = 0.50, meaning 0.50% in excess return for every 1.0% in risk.

No Diversification: Perfect Positive Correlation

Conversely, when the correlation is perfectly positive (1.00), the portfolio risk = 12.5% and the Sharpe Ratio = 0.36, meaning 0.36% in excess return for every 1.0% in risk. Note that 12.5% is precisely the average volatility of the two equally weighted asset classes.

Disclosure: I/we have no positions in any stocks mentioned, and no plans to initiate any positions within the next 72 hours.

I wrote this article myself, and it expresses my own opinions. I am not receiving compensation for it. I have no business relationship with any company whose stock is mentioned in this article.