Predicting Mutual Fund Performance Using (Legal) Inside Information

How does an investor choose which mutual fund to invest in? She’ll want a measure of the fund manager’s skill, and the most natural measure is his past performance. But, a ton of research has systematically found that past performance doesn’t predict future performance – it’s irrelevant in choosing a mutual fund.

How can this be? One interpretation is that fund managers aren’t skilled to begin with, and instead any good performance is due to luck. The thinking goes as follows. Skill is permanent. If good past performance were due to skill, performance should stay strong in the future. But, luck’s temporary.  If good past performance were due to luck, performance should revert to the average in the future. Since future performance appears unpredictable, this seems to support the luck explanation. This has huge implications for investors – if mutual fund managers indeed have no skill, there’s no point paying the high fees (around 1.5% per year) associated with actively-managed funds. Instead, put your money in passive index funds (where fees can be as low as 0.1%). Perhaps due to this thinking, passive index funds have grown substantially in recent years.

But an influential 2004 paper by Jonathan Berk (Stanford) and Rick Green (Carnegie Mellon) reached a different conclusion. Fund managers are skilled, and good past performance is a signal of skill. But, because everyone else is trying to invest with a skilled manager, managers with good past performance enjoy a flood of new funds coming in. This increases the fund manager’s assets under management (AuM) and thus his fees (which are a percentage of AuM) and so he won’t discourage the new flows. But, it will worsen his performance next year, because of diminishing returns to scale in investing. The manager has to put the new funds to work. But, he’s already investing in his top stock picks. He can’t put all of the new money in the same stocks, because there’s not enough liquidity in the market to accommodate this extra demand. So, he’ll have to choose his next-best picks, which will do worse. Thus, even though past performance is an indicator of skill, it’s not an indicator of future performance.

What’s the problem here? The analogy is choosing an individual stock. Choosing a stock on the basis of an attractive characteristic that’s known to everyone (e.g. buying Facebook because it’s a leader in social media) won’t be fruitful. Since everyone else is aware of that characteristic, they will have bought into the stock and driven the stock price up – the “Efficient Markets Hypothesis”. Similarly, identifying fund manager skill using a dimension that’s known to everyone (e.g. past performance) is also not fruitful. Since everyone else is aware of past performance, they will have bought into the fund and driven its AuM up, worsening its future performance.

The key to picking a stock is thus to identify positive attributes that might aren’t known to others. Similarly, the key to choosing a mutual fund is to find a measure of skill that isn’t known to others – to have a measure of skill based on private (but legal) inside information. This is where an ingenious new paper by Jonathan, together with Jules van Binsbergen (Wharton) and Binying Liu (Kellogg), entitled “Matching Capital and Labor”, comes in.

A mutual fund is part of a fund family. For example, the Fidelity South East Asia Fund and the Fidelity Low Priced Stock Fund are both part of Fidelity. One of Fidelity’s jobs as a fund family is to evaluate the performance of each fund manager, to decide whether to promote her (i.e. give her an additional fund to manage, or move her to a larger fund) or demote her (take away one of her funds). They have access to a ton of information over and above past performance figures – just like scouting out a baseball player gives you much more information than you’d get from the statistics. For example, they can engage in subjective evaluations of her performance based on on-the-job observation, or assess whether poor performance might actually be due to good long-run investments that just haven’t paid off yet.  Thus, a promotion signals positive private information, and a demotion signals negative private information.  

As an example, take Morris Smith. He joined Fidelity in 1982 and, from 1984-6, ran Fidelity's Select Leisure Fund, which soared from $500k to $350m under his management. In 1986 he was promoted to the Fidelity Over-the-Counter Fund and managed an average of $1b. After further good performance he was promoted to Fidelity's flagship fund in 1990 with assets of $13b.

In short, by observing promotion and demotion decisions (which we can, using data sources such as Morningstar and CRSP), we can infer the fund family’s private information.

Jonathan, Jules, and Binying find that:

1. Promotion and demotion decisions can’t be predicted using data on past performance. In other words, observing such decisions gives investors, additional information over and above what we’d get from past performance figures. It allows us to (legally) infer the fund family’s private information.
2. Promotion and demotion decisions both increase the fund manager’s value added.  The authors measure value added using a metric introduced by an earlier paper by Jonathan and Jules.  This equal’s the fund’s “gross alpha” (its actual return before fees and expenses, minus the return from passively holding the benchmark) multiplied by its assets under management (“AUM”).  This gives a dollar measure of how much value is added (or subtracted) by active management.  That both promotions and demotions increase future value added suggests that promotions give more capital to a skilled manager who can use it effectively, and demotions pull the plug from an unskilled manager who was using capital wastefully.  Thus, the information that promotion/demotion decisions give is not only incremental (to past performance), but also useful.
3. It’s inside information that drives the results.  “External” promotions or demotions (a manager leaving to a new fund family and managing a fund with higher or lower AUM than he did before) have no effect on future value added.
4. These effects are large. The fund family’s decision to promote or demote a manager adds value of $715,000 per manager per month. Thus, 30% of the value that a mutual fund manager adds comes from the fund family giving her the right amount of capital.

Why doesn’t the decision to give a manager a second fund lead to the problem in Berk/Green, that the fund manager now has too much money under her control? Because, the fund family – through its extensive monitoring – estimates the optimal amount of funds to give each manager. It chooses to promote managers who previously had been underallocated funds, so that promotion does not lead to the problem of diminishing returns to scale.

Decades of academic research have failed to find an answer to one of the most important practical questions for investors – how to predict mutual fund performance. Jonathan, Jules, and Binying may have just found a way.