I have to laugh whenever I hear someone still touting Modern Portfolio Theory as a preferred investment strategy. The notion seems to have eluded innovation over the past half century. Truth be told, much of the financial services world has ignored advances in portfolio risk management technology. Clinging to a tired theory as a viable means of managing client assets is the equivalent of making a 2010 trans-Atlantic flight on a propjet.
MPT may provide a useful framework for promoting asset allocation, but it fails to address the frequency and magnitude of extreme downside events and the damage they cause investor portfolios. Within the dynamic environment of investments, MPT is a static prescription for an outcome well into the future. But investors are concerned with their wealth and changing conditions in the interim, not just ten years from now.
Advances in portfolio theory and the computational power to quickly produce outcomes have overcome the limitations previously faced by the financial services industry. Why has the retail side of the business been so slow to adopt these new methods?
It might be that the industry is driven by productivity and MPT is easy to understand, though I would argue that some retail associates have a tough time calculating the return range for a one-standard-deviation event. When given proper implementation and disclosure, MPT has offered a low-liability template, at least so far. But using the term modern to describe a 50-year-old theory is a stretch at best.
MPT has known deficiencies, such as its use of standard deviation as a measure of risk. Another shortcoming is its application of normal distributions and mean variance optimization.
Normal distributions, the math behind MPT, fall apart in the non-normal distribution world of investment returns. Over long time periods, regular and irregular distributions of investment returns may be close enough to warrant the MPT model, but it fails miserably to model the frequency of extreme events over shorter time frames. Under normal distribution, 99.7% of returns take place within three standard deviations. Stock market history demonstrates that the negative returns of greater than three standard deviations occur far more frequently than the normal distribution math of MPT suggests.
One potential solution is Post-Modern Portfolio Theory (PMPT). This extends MPT by focusing on the downside risk-that is, on the return behavior of an investment below its target return rate. Think of downside risk as the annualized standard deviation of returns below the target rate. If an investment has a target return of 10%, any year when the return is below that number represents downside risk.
PMPT also addresses volatility distortion. MPT assumes returns are symmetrical with volatility distortion equaling 1.0 for all assets; that means returns are evenly distributed above and below the mean. PMPT identifies the distortion associated with a historical return pattern and assigns a distortion value. Values higher than 1.0 represent positive distortion and those below 1.0 represent negative distortion. By analyzing return distortion more closely, managers can construct portfolios with investments that tend to deliver greater positive distortion and lower downside risk.
Another more recent asset allocation model is called Dynamic Portfolio Optimization (DPO). This model promotes dynamic-risk-based investing, which contends that better risk methodologies and return forecasting models exist that allow portfolios to be more intelligently managed.
By defining a universe of global securities, typically ETFs and mutual funds, the investor can identify two or three favorable investments for a specific sector, industry or asset class. The filters include fees, price spreads, trading volume, tracking error, performance and other fundamentals. New candidates are continuously screened. Those likely to receive a small allocation or no allocation during the current period (usually monthly) are removed. Those remaining are subject to a series of simulation runs using univariate, bivariate and multivariate modeling to obtain the optimal asset allocation mix for the investment period.
The univariate step uses a non-normal "Student's t-distribution" to more accurately measure risk and identify outliers that would be missed with normal distribution methods. Then the GARCH model (generalized autoregressive conditional heteroskedasticity) is used to time-weight the data through cluster analysis. GARCH creates a Doppler radar effect, whereas traditional models using mean variance act like a Farmers' Almanac.
The bivariate process focuses on security correlation. Low correlation is considered your friend in asset allocation, that is until you need it most-during extreme events where correlation produces what is known in financial engineering as a "volatility smile." Correlation shortcomings are compensated for using an advanced method called copula dependency to identify the dynamic relationships between two securities as volatility and prices change.