How hard is it to beat equal weighting in constructing a portfolio of attractive common stocks? In his May 2016 paper entitled “Naive Diversification Isn’t so Naive after All”, Mike Dickson compares performances of 15 portfolio construction methods applied to eight portfolios of stocks with high expected returns. Construction methods include equal weighting, two versions of minimum volatility, three versions of mean-variance optimization, eight versions of reward-to-risk timing (six of which involve factor models) and a characteristic-based scheme that each year estimates stock weights based on market capitalization, book-to-market ratio, gross profitability, investment, short-term reversal and momentum. The eight portfolios consist of stocks with the top 10% or top 20% of expected returns based on rolling averages of multivariate cross-sectional regression coefficients for these same characteristics, formed with or without momentum and with or without microcaps (capitalizations less than the 20% percentile for NYSE stocks). He estimates trading frictions as 1% of the value traded each month in rebalancing to specified portfolio weights. Using monthly data for a broad sample of U.S. common stocks during July 1963 through December 2013 (with evaluated returns commencing July 1973), he finds that:
- For the full sample (including microcaps):
- Equal weighting has one of the largest net returns, largest Sharpe ratios and smallest turnovers of all portfolio construction methods.
- Minimum volatility with extreme estimator shrinkage is the strongest competitor for equal weighting due to its low volatility and low turnover.
- For the sample excluding microcaps, results reinforce the finding that equal weighting is hard to beat, with a Sharpe ratio close in value to those of all other top performing portfolio construction methods.
- Relatively low variation in individual stock Sharpe ratios within portfolios of top-performing stocks may explain why more sophisticated methods cannot reliably beat equal weighting.
- Results from Monte Carlo simulations confirm the relatively strong performance of equal weighting.
In summary, evidence indicates that equal weighting of stocks that are likely to outperform is hard to beat.
Cautions regarding findings include:
- Testing many strategies on the same underlying data sets introduces snooping bias, such that the best-performing strategy-data combinations impound luck and overstate expectations. Moreover, the strategies themselves have different numbers of parameters, such that snooping bias may affect some strategies more than others.
- Using a fixed estimate of trading frictions ignores differences across categories of stocks. For example, frictions are likely much higher for microcaps than for other stocks.
- Also, trading frictions vary considerably over the sample period such that the assumed level is likely too low early in the sample period and too high late in the sample period (see “Trading Frictions over the Long Run”). Since different strategies have different turnovers, checking relative performance by subperiod may be illuminating.
- As noted in the paper, double-use of the six stock characteristics to specify strategy weights and to pick stocks for the portfolios may explain the disappointing performance of the characteristic-based portfolio construction method.
