Skip to content

How Inefficient are Simple Asset-Allocation Strategies? (1/N): summary and key takeaways

The paper tests naive equal weighting against about ten static and dynamic optimizing portfolio models on nine empirical data sets plus simulations. The equal-weight rule usually matches or beats the optimizers out of sample on Sharpe ratio and turnover, because estimation error in the inputs offsets the gains from optimizing.

Garlappi, L.; DeMiguel, V.; Uppal, R. · 2005 · SSRNSummary updated 20 November 2026
Published2005Working paper
Asset focusAsset-agnosticPortfolio construction
TypePortfolio MgmtPortfolio Construction & Risk
Reading time4 minthis summary

How Inefficient are Simple Asset-Allocation Strategies? (1/N) at a glance

The question
How much does an investor lose by splitting wealth equally across assets instead of using the weights from optimal asset-allocation models?
Dataset & period
Nine empirical data sets, mostly monthly returns, in excess of the 90-day US T-bill. Examples: ten S&P sector portfolios (January 1981 to December 2002), ten industry portfolios (July 1963 to November 2004), nine international equity indexes, Fama-French size and book-to-market portfolios, and two quarterly stock-bond sets for dynamic models. Rolling estimation window of 120 observations. Plus simulated data.
Method
Each model is estimated on a rolling window, its weights are computed, and the portfolio is held for the next period, then the window moves forward. Strategies are compared on out-of-sample Sharpe ratio, certainty equivalent (CEQ) and turnover. Sharpe ratio differences are tested with the Jobson-Korkie test with the Memmel correction. Competitors include mean-variance, minimum-variance, Bayes-Stein shrinkage, a three-fund rule, a Data-and-Model Bayesian approach, short-sale constrained versions, and two dynamic models from Campbell and Viceira.

Key findings of How Inefficient are Simple Asset-Allocation Strategies? (1/N)

  • In the sector data, the mean-variance portfolio has the highest in-sample Sharpe ratio, 0.38481, but out of sample the two equal-weight rules have the highest Sharpe ratios; the rebalanced version is at 0.18762.
  • Across data sets, the unconstrained optimizing models (mean-variance, Bayes-Stein, three-fund, Data-and-Model) do much worse out of sample than the constrained ones and than equal weighting.
  • Constrained minimum-variance often posts the best out-of-sample Sharpe ratio, but its difference from equal weighting is typically not significant, and its turnover is 2-6 times higher.
  • Turnover in the sector data is 0.03051 for rebalanced equal weight and 1.1894 for the mean-variance strategy, the highest of all strategies.
  • In the size and book-to-market data under a single-factor model, constrained minimum-variance has the best Sharpe ratio, 0.25574, and the difference from equal weight is significant (p-values 0.07 and 0.04). Its CEQ is nonetheless lower than that of equal weight without rebalancing.
  • Dynamic strategies do not beat equal weight either. With stochastic interest rates the rebalanced equal-weight Sharpe ratio is 0.41105 against 0.35284 for the dynamic strategy, and dynamic policies often take extreme long and short positions.
  • In simulations, even with an estimation window of 100 years, the gap between equal weight and optimizing models is not statistically significant. In the base case, the Sharpe ratio is 0.13479 for rebalanced equal weight and 0.14079 for the single-asset and true-moment mean-variance strategies.

Robustness, as reported in the paper

  • Results are shown for nine empirical data sets plus simulated data, to avoid dependence on one sample.
  • The authors also ran a 60-observation estimation window for monthly data. A shorter window only hurts the optimizing models.
  • A one-year holding period, other risk-aversion levels and the portfolio including the riskless asset gave similar conclusions, per the paper.
  • The paper reports that it ran more than 360 tables and presents only the settings least favorable to equal weight.

Caveats and limits

  • Transaction costs are not modeled; turnover is shown only as a proxy for them.
  • The authors say the conclusion holds for the data sets and standard estimation methods used; better estimators could change the picture.
  • Equal weighting loses in simulations to true-moment optimization, as expected, so the result depends on how noisy the inputs are (inferred).
  • Sample ends in 2004 at the latest, so recent periods are not covered (inferred).

Items marked (inferred) are our reading of the design, not statements by the authors.

Why it matters

The paper shows that a no-estimation equal-weight portfolio is a hard benchmark for any optimizer. Practitioners can use it as a baseline to check whether the extra complexity of an optimized allocation actually adds out-of-sample value.

Questions about this paper

Why does equal weighting do so well out of sample?

The paper says the gain from optimal over naive diversification is smaller than the loss from estimation error in the inputs the optimizers need.

Does a longer estimation window fix the optimizers?

Performance of optimizing models improves with longer windows, but in simulations even with an estimation window of 100 years the difference from equal weight is not statistically significant.

Which optimizing strategy did best out of sample?

Usually constrained minimum-variance, but its Sharpe ratio and CEQ difference from equal weight was typically not significant, and its turnover was higher.

Where can I read the original paper?

The original is at its publisher: https://doi.org/10.2139/ssrn.676997. We link to the paper and do not host the PDF.

Follow new quant research

Download MacroFinance market data as CSV or Excel, or pull it by API, and test the ideas in these papers yourself. New summaries are added as we review them.

Create a free account
Summaries are written by MacroFinance from the published papers, in our own words, for information only. They are not investment advice and may omit detail. Papers are not hosted here: each page links to the original at its publisher.