A Dollar of Stocks for Eighty Cents: The Closed-End Fund Puzzle

Sentiment is behavioral finance's most frustrating concept. Everyone agrees investors have moods. Nobody can measure them.

You cannot poll every investor. Surveys are unreliable. And if you just use past returns as a proxy for sentiment, you are arguing in a circle. So the whole edifice of noise trader theory (De Long, Shleifer, Summers and Waldmann in 1990) rests on a variable nobody could observe.

Lee, Shleifer and Thaler found one. And the beauty of it is that they did not invent a new statistic. They pointed at a puzzle that had been sitting in plain sight, annoying economists for decades, and said: that is not a puzzle, that is a thermometer.

The problem: a fund worth less than the sum of its parts

A closed-end fund is a simple thing. It raises money once, buys a portfolio of stocks, and then its own shares trade on an exchange. Unlike an ordinary mutual fund or an ETF, it does not continuously create and redeem shares. It is a fixed pot.

So a closed-end fund has two prices, and this is what makes it such a gift to researchers:

The net asset value (NAV): just add up the market value of the stocks it holds. This is not an estimate. These are listed stocks with observable prices. You can compute it to the penny.
The fund's own share price: whatever the fund itself trades at on the exchange.

Basic logic says these should be the same. If a fund holds 100 dollars of Apple and Microsoft, a share of that fund is a claim on 100 dollars of Apple and Microsoft. Anything else is odd.

In reality, closed-end funds typically trade at a discount, often 10 to 20 percent below their net asset value. You can buy a dollar of blue-chip stocks for eighty cents. And this is not an obscure micro-cap phenomenon, it is the normal, persistent state of a well-known asset class, and it had been embarrassing economists since the 1930s.

Worse, the puzzle has four parts, and no rational explanation handles all of them:

Funds usually start out trading at a premium (people pay more than NAV at the IPO), then slide to a discount within months.
The discounts vary enormously over time, swinging around far more than any fee or tax story can explain.
When a fund is liquidated or converted to an open-end structure, the discount vanishes and the price snaps to NAV. So the discount was never justified by anything real about the assets.
Discounts across completely different funds holding completely different stocks all move together.

That last one is the smoking gun.

The key idea via analogy: reading the room's mood

Suppose you notice that in a hundred unrelated offices across a city, the thermostats are all being nudged up and down in unison. The offices have different buildings, different insulation, different equipment. There is no mechanical reason for them to be synchronized.

The only sensible explanation is that something common is acting on all of them: the weather outside, or perhaps a shared mood.

That is Lee, Shleifer and Thaler's argument. If closed-end fund discounts were caused by fund-specific factors (bad managers, high fees, illiquid holdings, embedded tax liabilities), then each fund's discount should follow its own path. A fund with a terrible manager should have a big discount and a fund with a good one should not, and the two should have nothing to do with each other.

Instead, the discounts move together. Something is acting on all of them simultaneously. And that something is what the authors call investor sentiment.

Here is the mechanism. Closed-end funds in this era were held overwhelmingly by individual investors, not institutions. So the price of a closed-end fund reflects, to an unusual degree, the mood of retail investors specifically. When individuals are optimistic, they bid up closed-end funds and the discount narrows. When they turn gloomy, they sell, and the discount widens. The discount is a mood ring for the small investor.

And now the paper delivers its knockout test. If the discount really measures retail sentiment, it should predict the returns of other assets that retail investors dominate. Which assets are those? Small-cap stocks, which institutions largely ignored and individuals loved.

So the prediction: closed-end fund discounts should move with small-cap stock returns, even though the funds themselves mostly hold large-cap stocks, and there is no mechanical link whatsoever between them.

They tested it. It held. When discounts narrowed (retail optimism), small stocks did well. When discounts widened (retail gloom), small stocks did badly.

Think about how strange this is under a rational model. The discount on a fund holding IBM and General Electric should have nothing to do with the returns of tiny companies the fund does not own. The only thing connecting them is the mood of the people who hold both. That connection is exactly what noise trader theory predicts and exactly what rational asset pricing forbids.

Why it mattered

It gave sentiment a number. Before this, "investor sentiment" was hand-waving. After this, it was a time series you could regress things on. The entire modern sentiment literature (Baker and Wurgler's sentiment index and its many descendants) descends from this paper's core move: find an asset whose price reveals the mood of a particular investor clientele, and use it as a thermometer.
It is the empirical companion to De Long, Shleifer, Summers and Waldmann. That 1990 paper argued that noise trader mood is a priced risk: rational arbitrageurs will not close a gap if the noise traders' mood might get worse before it gets better. The closed-end fund discount is precisely that. An arbitrageur could buy the discounted fund and short its holdings, but the discount could widen first, and he might be carried out before it closes. So the discount persists. Theory in 1990, evidence in 1991.
It is one of the cleanest mispricings ever documented. Most claimed anomalies require a risk model, and are therefore arguable. Here, you can see both prices. There is no model. The fund is worth less than the things inside it, and everyone can check.
It reframes a nuisance as a signal. The best papers do this: take something the profession had filed under "annoying anomaly, ignore it" and reveal it as a window into how markets actually work.

The honest limitations

The rational camp fought back hard. There are respectable rational explanations for parts of the puzzle: management fees drag on value, embedded capital gains create a future tax liability for the buyer, and some funds hold genuinely illiquid assets whose marked value is optimistic. The counter-argument is that none of these vary enough over time, or correlate enough across funds, to produce the co-movement. But the debate is real, and it has never fully closed.
The small-cap correlation could run the other way. Maybe small-cap returns drive both retail mood and fund discounts, rather than sentiment driving both. Causality here is genuinely murky, and the paper's evidence is correlational.
Closed-end funds have shrunk in importance. In 1991 they were a significant retail vehicle. Today, ETFs (which have a creation and redemption mechanism that keeps price glued to NAV, precisely because someone can arbitrage it) have largely replaced them. The thermometer still exists but the patient has moved. Ironically, the fact that ETFs don't have persistent discounts is itself the strongest possible evidence for the paper's thesis: give arbitrageurs a riskless mechanism and the discount disappears instantly.
Sentiment remains slippery. Even granting everything, "sentiment" is defined by its effects rather than measured directly. This paper narrows the gap enormously. It does not eliminate it.

The one-line takeaway

Lee, Shleifer and Thaler showed that closed-end funds trade at persistent discounts to the stocks they hold, that those discounts all move together across totally unrelated funds, and that they track the returns of small-cap stocks that retail investors love, which means the discount is not a valuation error to be explained away but a working thermometer for the mood of the small investor.