Luck vs Skill in Investing

Quick definition: The skill-versus-luck framework in investing uses statistical methods to determine whether an active manager's outperformance is likely due to genuine stock-picking skill or simply due to random variation (luck) that could occur by chance alone.

Key Takeaways

• A manager would need to outperform by 2–3 percentage points annually for approximately 15–20 years to accumulate statistical evidence that the outperformance is due to skill rather than luck.
• Using an information ratio (excess return divided by tracking error), a manager needs an information ratio above 1.0 consistently for 5+ years to be considered likely to possess genuine skill.
• With approximately 5,000 active managers in the US equity market, random chance alone would produce approximately 50 "lucky" managers with 10-year track records appearing to outperform by 2+ percentage points.
• The "p-hacking" problem in manager selection: if you test enough managers across enough metrics, you will find some who appear to have skill purely by chance.
• Even among legendary investors like Warren Buffett, rigorous statistical analysis suggests that while their returns are unlikely to be purely random, the magnitude of skill is smaller than commonly believed.

The Statistical Framework: How Much Outperformance Is Skill?

The ability to distinguish luck from skill in investing rests on understanding hypothesis testing and statistical power. The basic question is: given a manager's observed outperformance, what is the probability that the outperformance is due to skill versus luck?

To answer this question rigorously, we need to quantify:

1. The magnitude of outperformance (e.g., +2% annually)
2. The volatility of returns (tracking error around the benchmark)
3. The length of the track record (5 years, 10 years, 15 years)

From these inputs, we can calculate the information ratio: the excess return divided by the tracking error. An information ratio of 0.5 means the manager beat the benchmark by 0.5 units of standard deviation (one unit of tracking error per year). An information ratio of 1.0 means the manager beat the benchmark by 1 standard deviation. An information ratio of 2.0 is extremely high—it would imply a very unlikely outcome under random variation.

The Benchmark Calculation: How Many Years to Prove Skill?

Research by Larry Summers and others has estimated that to have 95% confidence that a manager's outperformance is due to skill (and only 5% probability it is due to luck), a manager would need to accumulate one of the following:

• 3% annual outperformance over 20 years (cumulative outperformance of 60%)
• 2.5% annual outperformance over 25 years (cumulative outperformance of 62.5%)
• 2% annual outperformance over 30 years (cumulative outperformance of 60%)

Alternatively, using information ratio thresholds:

• Information ratio of 1.0 over 5 years is borderline evidence of skill
• Information ratio of 1.5 over 5 years is stronger evidence of skill
• Information ratio of 2.0 or higher over 5 years is very strong evidence of skill

Most active managers have information ratios of 0.0–0.5, meaning they underperform or barely match the index after accounting for tracking error. To have 95% confidence of skill, a manager needs an information ratio above 1.0, which is extremely rare.

Empirical Study: How Rare Is Genuine Skill?

A study by Kosowski, Naik, and Teo (2007) examined the performance of hedge funds (which charge much higher fees, supposedly allowing more room for manager skill) to estimate how many genuinely skilled managers exist. They used Bayesian analysis to estimate what fraction of the observed distribution of fund returns would be expected if all managers had zero skill, versus what fraction could be attributed to genuine skill.

Their findings:

• Among 1,500+ hedge funds analyzed, if all managers had zero skill, approximately 75 funds would appear to have 10-year track records beating the market by 5%+ annually purely by chance.
• The actual number of funds appearing to have such strong performance was approximately 140.
• This suggests that about 65 hedge funds (140 minus 75) have genuine skill, or roughly 4% of the population.
• However, these skilled managers did not persist: the skilled managers of decade one were not the same as the skilled managers of decade two.

This study is crucial because hedge funds have far fewer regulatory constraints than mutual funds and charge much higher fees, creating more room for manager skill to be expressed. If only 4% of hedge funds have genuine skill, the fraction for mutual funds (with more constraints and regulatory oversight) is likely lower, perhaps 1–3%.

The Problem of Multiple Testing: Why Randomness Creates Winners

Imagine an experiment: 5,000 active US equity managers flip coins every year to decide whether to beat or underperform the S&P 500 by 1 percentage point. On average, 2,500 would beat the index in year one, 1,250 in year two, and so on. After 10 years, you would expect approximately 5 managers to beat the index by 1%+ annually for all 10 years purely by chance (5,000 times 0.5^10).

In reality, there are approximately 5,000–6,000 active US equity mutual funds and separately managed accounts in operation at any given time. Even if all of them had zero skill and were essentially flipping coins, random variation would produce a handful of managers who appear to have beaten the market for 10+ years purely by chance.

This is the multiple-testing problem: when you test enough managers, you will inevitably find some who look like winners purely by luck. The solution is to apply a statistical correction (like Bonferroni correction) that adjusts the significance threshold based on the number of managers being tested. When these corrections are applied, most managers' apparent outperformance becomes statistically insignificant.

Information Ratio in Practice: What Do Real Managers Look Like?

Let us examine what real information ratios look like for active managers:

Typical large-cap active manager:

• Annualized excess return: +0.5% (after fees, versus S&P 500)
• Annualized tracking error: 3–4%
• Information ratio: 0.5/3.5 = 0.14

Top-quartile large-cap active manager:

• Annualized excess return: +1.5% (after fees)
• Annualized tracking error: 4–5%
• Information ratio: 1.5/4.5 = 0.33

Rare outperforming manager (percentile <5):

• Annualized excess return: +2.5% (after fees)
• Annualized tracking error: 5–6%
• Information ratio: 2.5/5.5 = 0.45

Even the best active managers in the real world have information ratios of 0.4–0.6 over 10–20 year periods. To have statistical confidence in skill (information ratio >1.0), you need to observe margins of outperformance that are extremely rare in practice. This mathematical reality implies that even if some managers do have genuine skill, identifying them in advance is nearly impossible.

Case Study: Warren Buffett and the Skill-Luck Question

Warren Buffett is the most commonly cited example of an active manager with apparent genuine skill. His Berkshire Hathaway outperformed the S&P 500 by approximately 3.4% annually over 60 years (1965–2023), with an information ratio of roughly 0.7–0.9. Over a 60-year period, an information ratio of 0.7 translates to a t-statistic of approximately 5.5, which has a p-value of roughly 0.0001—extremely unlikely to occur by chance alone.

However, even Buffett's remarkable record involves important caveats:

1. Survivor bias in manager selection: Buffett got into investing at the right time, in the right market, with the right personal circumstances. There were many other brilliant investors in the 1950s–1960s who are now forgotten because they made different bets or had bad luck. Buffett was lucky in ways we cannot easily quantify.

2. Factor exposure: Buffett's portfolio has had consistent exposure to value factors, which outperformed during much of his track record. Some of his outperformance may be due to being long value stocks (which tend to outperform over long periods), not pure stock-picking skill.

3. Scale issues: Buffett's outperformance has declined as Berkshire's size has increased. A $1 trillion company has difficulty beating the index, whereas a $100 million company can more easily beat the market through concentrated positions and nimble trading. As Buffett's assets under management grew, his alpha shrank.

4. Time period specificity: Buffett's best years were the 1970s–1980s, when value investing was deeply out of favor. A manager who bets heavily on an unloved factor for 20 years and that factor then outperforms for 20 years gets credit for "skill," even though half the result was luck.

Even adjusting for these factors, Buffett likely has genuine skill in stock selection and capital allocation. But the magnitude of his edge, once these adjustments are made, is smaller than the headline numbers suggest—perhaps 1–2% annually rather than 3.4%. And identifying another Buffett in advance is nearly impossible, because Buffett's success involved idiosyncratic factors (his network, his personality, his contrarian bets, his very long time horizon) that are difficult for other managers to replicate.

The Impossibility of Identification: The Central Problem

Here is the core paradox: even if skilled managers exist, identifying them in advance is nearly impossible because:

1. The skill signal is weak: An actual skilled manager might have an information ratio of 0.5–0.7, which is below the threshold (>1.0) needed for high statistical confidence.

2. Noise dominates: In any given 5–10 year period, random variation can easily produce apparent information ratios of 0.4–0.6, masking the true information ratio of a skilled manager and making him indistinguishable from a lucky one.

3. Multiple testing problem: With 5,000 managers to choose from, random chance alone creates 50–100 managers who look like outliers just by luck.

4. Regression to the mean: A manager with an apparent information ratio of 0.6 in one decade will likely have an information ratio of 0.3–0.4 in the next decade, as the lucky component of returns reverses.

This is why performance chasing—selecting a manager based on recent outperformance—is so dangerous. The investor is not identifying skill; they are identifying luck that is about to reverse.

Bayesian Perspective: Prior Probability Matters

Another way to think about skill versus luck is through Bayes' theorem. Suppose you have a prior belief about the probability that a manager has genuine skill (say, 5%, based on the Kosowski study). You observe that the manager beats the benchmark by 2% over 5 years (information ratio of 0.4). Does this observation increase your posterior probability that the manager has skill?

Using Bayesian updating, the answer is: not by as much as you might think. The 2% observation is consistent with both luck and modest skill, so it does not drastically change the posterior probability from the prior.

Only if you observed the manager beat the benchmark by 5–6% over 5 years (information ratio of 1.0+) would the posterior probability of genuine skill increase significantly. But such outcomes are extremely rare and may reflect selection bias (you are looking at a manager you have already chosen to follow, not a randomly selected manager).

Conclusion: The Skill-Luck Boundary in Practice

The mathematical evidence is clear: distinguishing luck from skill in active management requires either (a) extraordinarily large margins of outperformance (5–6% annually), or (b) extremely long track records (20–30 years). Most active managers do not meet either criterion. This does not mean all active managers are unskilled, but rather that the signal of skill is too weak to detect reliably and predict in advance.

For investors, the practical implication is that the odds of picking a skilled manager in advance are low—lower than the odds of simply investing in a passive index and avoiding the problem of selection altogether.

Decision tree

Next

Next, we quantify the drag on active management returns caused by fees, examining historical fee trends and the cumulative impact of a seemingly small 1% annual fee compounded over decades.