What True Diversification Really Is And How To Maximize It


Key Takeaways:

Maximize reward per risk
  • To do this we maintain a diverse exposure to various edges
    • Edges are forms of risk premia: ie value, momentum, quality.
    • Is a manager who is constrained to allocating, or applying their edge in the asset allocation space able to produce the same risk adjusted performance as a manager who's operating in the security selection space, given that there's far fewer asset classes than there are securities?
    • We want to diversify sources of risk premia as opposed to simply diversifying securities. A basket of 500 stocks has less diversity than exposure to 10 different asset classes.
    • Ideally, you want to apply forms of edges to a wide set of uncorrelated bets on meaningful time frames. That represents the opportunity set.
Measure of diversification: Unique uncorrelated bets
  • They did an analysis of Ken French equity data going back 100 years to study the performance of 10 industries. Accounting for correlations this reduced to only 1.5 unique bets with the rest of the risk greatly overlapping. If you the same analysis on the 38 industries as defined by S&P you still only get 3 unique bets.
  • When they do this same analysis across 12 different asset classes (ie US stocks, Asia stocks, emerging stocks, treasuries, bonds of different durations and credit, gold) you get 5 unique or independent bets. Compare this with how 38 US industries only reduced to 3 bets.
  • As they extrapolate the analysis, it turns out that it takes 500 stocks to give us the same number of unique bets as 12 asset classes…the problem is the intra-cross correlation of the stocks tends to increase rapidly in episodes of market stress.
    • When we're looking at 500 stocks, we're in the business of picking a better tree in a forest. Which tree is looking better from a momentum perspective? Whereas asset allocation is in the business of picking better forests. In 2008 the S&P 500 forest was burning down and you could have picked the best tree in that burning forest and not done so well. You'd rather get to that helicopter and start choosing forests that are thriving.
  • The desired diversification is actually diversification across real underlying economic factors. Different assets perform differently depending on economic circumstances. When we zoom out and look at quarterly or yearly data, we see the behavior of various asset classes in response to economic conditions is durable, even though on a daily basis there is noise or short term correlations can increase.
    • In 2008, it’s true that the basket of 500 stocks collapsed from 12 bets to 1 as correlations converged to 1. This absolutely does not happen at the asset class level. The dispersion between asset classes actually reached decade-wide disparities. The stock bond correlation actually went negative.
  • 48 futures markets can reduce into a healthy 12 or 13 unique bets to apply an edge algorithm to. Increasing the diversification to this degree can improve the Sharpe by 1/3, which means if you increase the bet size so that your diversified portfolio has the same volatility as your less diversified portfolio, you can grab an extra 1/3 return. If expected returns are 6%, that means an extra 2% annually. (This helps me think about how much we might be willing to pay for a strategy that can be exposed to this).