Sharpe's style analysis: what a fund really does

Advanced9 min read · updated 4 September 2026 · 6 sources

William Sharpe's style analysis (1992) answers a simple question using nothing but return data: which mix of style indexes (such as large value stocks, small caps or bonds) would have best replicated a fund's returns? The rest is what the managers contributed themselves. You do not need holdings lists, just monthly returns for the fund and a few benchmark indexes.

How it works

The goal is to find weights for a set of style indexes that add up to 100%, are not negative and minimize the deviation from the fund's return. Unlike an ordinary regression, this produces genuine portfolio weights that you could buy with index funds. The weighted mix is called the style benchmark, and the difference from the fund's return is called the selection return.

Mathematically, this is a quadratic optimization problem: the objective function depends on the covariance matrix of the style indexes and their covariances with the fund. Without the two constraints, it would be an ordinary regression. But it is precisely the constraints that make the weights easy to interpret: a 20% weight in small caps means the fund behaved like a portfolio with one fifth in small caps.

How to read R² and tracking error

The R² tells you what share of the fund's fluctuations the style mix explains. It is directly linked to the tracking error: R² = 1 − (tracking error / volatility of the fund)².

For broad stock funds, R² is usually above 0.9. A markedly lower R² means that an important style index is missing or that the fund pursues an independent strategy with concentrated bets, derivatives or leverage. The selection return alone says little; only together with the tracking error does it give the information ratio of the selection (see tracking error).

Style drift and out-of-sample testing

If you estimate the weights on a rolling basis over a window of about three years, you can see how the style shifts over time. A “value fund” that moves heavily into growth stocks along the way exhibits style drift. That can be deliberate tactics or a sign that the mandate is not being followed.

For a fair measure of selection, Sharpe (1992) recommended building the style benchmark for each month using only weights estimated from the preceding months. Otherwise, you fit the benchmark after the fact so that it matches the fund as closely as possible, and you underestimate the fund's own contribution. The out-of-sample selection return is therefore the more honest number.

• Spotting style drift: look at the rolling weights over time.
• A suitable benchmark: the style mix is fairer than a single index because it reflects the fund's actual orientation.
• Style map: with four corners (large value, large growth, small value, small growth), the style can be shown as a point on a map. It resembles a style box, but built from returns instead of holdings.

What research shows

Sharpe (1992) introduced the method and applied it to US mutual funds. He used a set of asset class indexes, ranging from money market and bonds of various maturities to stocks by size, style and region. The approach builds on the question of how much of the return is explained by the allocation across asset classes. Brinson et al. (1986) had broken down pension fund returns into investment policy, market timing and security selection; the strategic allocation explained most of the variation over time.

Ibbotson and Kaplan (2000) clarified that it depends on the question: investment policy explains about 90% of the variation of a typical fund over time, but only about 40% of the differences between funds, and on average about 100% of the level of return. Style analysis makes exactly this investment policy visible from returns.

Lobosco and DiBartolomeo (1997) were the first to derive confidence intervals for the style weights and tested them with Monte Carlo simulations. Daniel et al. (1997) describe the counterpart to returns-based analysis: they compare each stock in the fund with passive portfolios of the same size, valuation and prior-year return. They applied this holdings-based method to more than 2,500 stock funds from 1975 to 1994.

Advanced: how precise are style weights?

Style weights are estimates, too. The approximation by Lobosco and DiBartolomeo (1997) for the standard error of a weight is:

• Similar indexes make weights uncertain: the more an index is explained by the others (high R²_i), the larger the standard error. Two nearly identical indexes split the weight between them arbitrarily.
• Weights at the boundary: if a weight is zero, there is no meaningful standard error for it, because the constraint is binding.
• More data helps: longer periods reduce the standard error with the square root of the degrees of freedom; Lobosco and DiBartolomeo (1997) also mention daily data as a way to narrower intervals.
• Attribution: because the style mix is a linear combination, the fund's average return breaks down exactly into weight × average return for each style, plus selection.

How to do it in practice

1. Define the question: should the analysis show the allocation across asset classes (balanced funds), the regions (global stock funds) or the style within a market (value, growth, size)? The choice of indexes depends on it.
2. Choose the indexes: four to eight clearly distinct, investable indexes in the same currency as the fund, plus a money market index for cash.
3. Run the full-period analysis: look at the weights, standard errors and R² over the entire common period. Do not read too much into weights with large standard errors.
4. Check on a rolling basis: if the weights shift markedly, a single full-period result means little.
5. Evaluate selection: look at the out-of-sample selection return, its t-statistic and the information ratio, and compare them with the fund's expense ratio.

A typical result for expensive active funds: the style mix explains almost everything, and the selection return is negative by roughly the amount of the costs. That fits the reasoning of Sharpe (1991), according to which the average actively managed dollar must lag the passive dollar after costs.

Limitations and common mistakes

The analysis sees only what shows up in returns. Funds that use derivatives, leverage or concentrated bets are hard to describe (low R²). The choice of style indexes determines the result: if an important building block is missing, its influence ends up in the other weights. A fund holding emerging market stocks that is analyzed with only US and European indexes gets distorted weights and a spurious selection return.

• Too many indexes: many highly correlated indexes make the weights unstable; four to eight clearly distinct ones are usually enough.
• Currency: calculate the fund and the indexes in the same currency; otherwise the exchange rate explains part of the deviation.
• Short windows: rolling weights from 24 months fluctuate a lot; not every move is style drift.
• Luck or skill: a positive selection return over five years is rarely statistically significant; the t-statistic of the selection shows how reliable it is.

As a complement, factor regression measures exposures to long-short factors without weight constraints. What the fund actually holds is shown by a look-through of its holdings lists.

How Portfolio Lab calculates

The Style analysis tool in Portfolio Lab solves the quadratic problem exactly, either in Sharpe's classic form (weights ≥ 0, sum 100%), with negative weights allowed, or without any constraints. There are templates for asset classes in euros (modeled on Sharpe's study), equity regions, bonds by maturity and US equity styles with a style map; you can also make your own selection of up to 15 indexes. The calculation uses completed months, at least 12 and at least three more than there are indexes. The output includes weights with standard errors following Lobosco and DiBartolomeo, R², selection per year (the difference between the annual growth rates), t-statistic, tracking error and information ratio. The rolling analysis uses 24 to 120 months (36 by default); the out-of-sample style benchmark uses, for each month, the weights up to the previous month. Details are in the style analysis methodology.

This article explains a method; it is not investment or tax advice. Worked examples are simplified and rounded. All terms in the glossary.