Once an investment product has been designed, we move on to its implementation. In this stage, we develop a series of rules and laws to govern portfolio behavior in order to service the portfolio goals and targets laid out in the design phase.
When we develop portfolio rules, we do it independently of performance testing, to the greatest extent possible, so that we are not biased by hypothetical results.
We also strive to implement a portfolio design with the fewest number of rules possible. By doing so, we reduce the overall complexity of the product, limit the number of “moving parts,” enforce robustness and generality in each rule we create, and ultimately allow for better understanding of the product by financial advisors and end investors.
The rules we develop have three primary characteristics.
Firstly, and most importantly, the rules should be as simple as possible. When comparing multiple possible rules that target the same goal, we employ Occam’s razor, always choosing the simplest solution. By reducing the complexity of a rule to its simplest possible implementation, we reduce the risk of over-optimization and hindsight bias.
Over-optimization is the risk that our rules have been optimally designed for specific situations, but will not work going forward. Overly complex rules that are prone to over-optimization are often the result of hindsight bias. Hindsight bias is the risk that we develop rules that are influenced by our knowledge of how events occurred. Unchecked, these risks will actually manifest themselves as a more performant back-test, but are almost guaranteed to break going forward. For a rule to be robust to changing markets, its implementation must be as simple as possible.
Secondly, rules should be adaptive; we consider fixed values (or “magic numbers”) and processes to be brittle foundations to build a portfolio on. For any rule governing portfolio behavior to remain relevant, it should adapt to changing market environments.
For example, pre-2008, the maximum level of the CBOE’s S&P 500 Implied Volatility Index (“the VIX”) was in the 30s. If we used the VIX as a measure of market risk in one of our rules governing portfolio allocation, and the process was not adaptive, our models would break as the VIX spiked to well above 80 in 2008.
However, our cardinal rule of simplicity must not be broken in the pursuit of rule adaptability; we strive to make the adaptive process as simple as possible as well.
Finally, portfolio rules must also be “philosophically defensible,” forcing us to idenfity and justify the assumptions that each rule stands upon. This prevents not only the inclusion of arbitrary rules (often due to hindsight bias), but also prevents over-optimization. If a rule is no longer recognizable or understandable from an economic, financial, or market interpretation, it means that it will likely break going forward. By justifying each rule, we are ultimately forced to identify the assumptions that the entire portfolio is being built upon, giving us the opportunity to recognize not only if we disagree with them, but also if any of the assumptions contradict each other.
As an example, let’s say we used the level of the VIX as a measure of risk to drive our overall bullish or bearish portfolio bias. This seems reasonable at first glance, but taking a step back to the definition of the VIX (annualized 30-day implied volatility on the S&P 500), it becomes harder to justify why a given volatility level necessarily implies a bullish or bearish stance. Empirically, however, it works. Clusters of higher volatility typically coincide with negative market returns; behavioral finance would defend this phenomenon by claiming that uncertainty, manifested and measured in high volatility, leads to herding behavior and inefficiency in the market place. By forcing ourselves to explore the justification for this rule, we discover the assumption from which the rule is built, and ultimately force ourselves to ask whether we are comfortable basing a portfolio on the assumption or not.
While there are always exceptions, this combination of characteristics makes the rules we develop robust to changing markets. The purpose of these framework characteristics is to mitigate, and hopefully prevent, the risks of complexity, over-optimization, unidentified assumptions, and hindsight bias.