Small-Cap Tilt with a Quality Screen

Overview

The size premium is the oldest embarrassment in factor investing. Banz documented it in 1981: small companies outperform large ones. It went into the Fama-French three-factor model as SMB, small minus big. It became gospel.

Then it stopped working. From the 1980s onwards, the raw size premium in the US has been statistically indistinguishable from zero. Small caps have been more volatile, less liquid and no better rewarded. A generation of investors who tilted small on the strength of the academic evidence got nothing for their trouble.

The interesting part came later. Asness and co-authors showed that the size premium does not really disappear, it gets buried. Small caps contain an enormous quantity of genuinely bad companies: unprofitable, over-levered, cash-burning, speculative shells. Large caps contain far fewer, because a company generally has to be doing something right to get large. So when you compare small to large, you are unknowingly also comparing junk to quality, and the junk drags the small-cap return down to nothing.

Control for quality, and a real size premium reappears. Small, good companies do beat large, good companies.

Thesis (why the edge exists)

If there is a size premium, why would it be there?

Illiquidity compensation. Small caps are harder to trade. You cannot get out fast, spreads are wide, and impact is real. Investors demand extra return for accepting that. This is a genuine risk premium and it should persist, but note carefully: it is a premium you earn by bearing a cost, and if your trading is sloppy you will pay the cost without collecting the premium.

Neglect and coverage. Fewer analysts follow small companies. Less information is priced in. Mispricings survive longer. This is plausible but weakening, since data is cheap now and quant funds cover everything.

Limits to arbitrage. Big institutions literally cannot own meaningful positions in tiny companies without becoming the market. That leaves the space to smaller players and lets mispricing persist.

The honest position: the raw premium is weak and possibly gone. The quality-screened version has better evidence, but part of what you are being paid for is illiquidity, which means the strategy has a hard capacity ceiling and punishes bad execution ruthlessly.

Strategy logic

• Cut the bottom off. Before you do anything else, remove the names that are not really investable: sub-penny prices, near-zero daily volume, shell companies. This is not cheating, it is refusing to pretend you could trade something you could not.
• Rank on size. Market capitalisation, ascending. Take the smaller portion of the universe.
• Screen for quality. Require the company to actually make money. Positive gross profitability, positive operating cash flow, leverage that is not extreme. Optionally add a low-accruals requirement so the earnings are real.
• Build the portfolio. Liquidity-weighted, capped, diversified across many names.
• Rebalance slowly. Every trade in small caps costs real money.

Parameters (knobs)

• Size cutoff: bottom 30 percent by market cap, or bottom half, or the classic Russell 2000 style band. Going deeper into micro-caps increases the theoretical premium and increases the practical impossibility of harvesting it.
• Quality screen strictness: a light screen (just positive gross profit) removes the worst offenders. A heavy screen (full quality composite in the top half) removes more but also cuts the universe substantially.
• Liquidity floor: the single most important parameter for whether this strategy is real or fiction. Set it based on your actual assets under management.
• Rebalance frequency: quarterly at most. Semi-annual is often better once costs are modelled.
• Weighting: equal weight (academic, overweights the least tradeable names), liquidity weight (practical), or capped market-cap weight within the small universe.

Portfolio construction

Diversify aggressively. Small-cap single names have fat tails: fraud, dilution, going concern, a lost customer that was 40 percent of revenue. Hold hundreds of positions, not dozens, and cap every one of them.

Weight by liquidity. Equal weighting a small-cap book means putting the same money into a company that trades 50 million a day and one that trades 200 thousand a day, which is how backtests generate returns that cannot be captured.

Sector-neutralise if you can. Small-cap universes swing wildly in sector composition, with biotech and regional banks dominating in ways that have nothing to do with the size premium.

Costs, capacity and turnover

This is the section that decides whether the strategy exists for you.

Spreads in small caps are wide, sometimes 30 to 100 basis points or more. Market impact is significant: if you need to buy 10 percent of a stock's daily volume, you will move the price against yourself. And critically, the cost is asymmetric, because when you want to sell in a stressed market the liquidity you assumed simply is not there.

Capacity is the binding constraint, full stop. A small-cap strategy that works at 10 million dollars may be entirely dead at 500 million, not because the signal decayed but because you have become the market. Compute your capacity ceiling before you compute your Sharpe ratio.

Turnover should be low by design. If your backtest has a small-cap strategy trading monthly with 200 percent turnover and still showing great returns, your cost model is broken.

Backtest design checklist

• Survivorship and delisting. Small-cap databases are riddled with this problem, and small caps are exactly where companies go to die. Missing delistings can inflate returns by several percent a year.
• Delisting returns. When a company is delisted, the investor does not get out at the last quoted price. Model a realistic delisting return, often deeply negative.
• Liquidity-aware execution. Cap your simulated trades at a fixed percentage of historical daily volume. If the backtest wants to buy more than that, it does not get to.
• Realistic spreads. Do not use a flat 10 basis points. Estimate spread from the data, by name, and let it widen in stressed periods.
• The unscreened comparison. Always run the raw small-cap portfolio alongside. The gap between it and your screened version is the entire thesis of the strategy, and if the gap is small, you do not have a strategy.
• Post-1980 sample. If your backtest's size premium comes mostly from the 1970s, you have discovered history, not an edge.

Common failure modes

• The capacity wall. The strategy works, right up until you have enough money for it to matter, and then it does not.
• Crisis behaviour. Small caps fall harder and recover later. In 2008 and 2020 the liquidity in small caps evaporated precisely when investors wanted to sell.
• Junk contamination. If the quality screen is too loose, you have rebuilt the original zero-premium portfolio.
• Data quality. Small-cap fundamentals are less reliable, filed later and restated more often.
• Idiosyncratic disasters. Single-name blowups are far more common down here. Fraud, dilutive equity raises, key-customer loss. Diversification is not optional.
• Fake backtests. More small-cap strategies die on execution than on signal. The backtest was never real.

Variants

• Small-cap value with a quality screen. Combining size, value and quality is the classic three-way tilt and has the strongest evidence of any variant.
• Small-cap momentum. Momentum is stronger in small caps than large, but so are the costs. The net is often disappointing.
• Micro-cap version. Higher theoretical premium, near-zero real capacity. Interesting for personal accounts, useless for a fund.
• Long small quality, short small junk. A pure expression that hedges out the small-cap beta, though the borrow on small-cap junk is often prohibitive.
• Equal-weight index tilt. The laziest version: buy an equal-weighted index rather than a cap-weighted one, which mechanically tilts small. Cheap, liquid, and captures a fraction of the effect.

Our notes and suggestions

The lesson of this strategy is larger than the strategy. The size premium was the textbook example of a factor, taught for forty years, and the raw version has not worked for most of that time. The reason it "came back" is that someone controlled for a confound that had been sitting in plain sight.

So build both portfolios, screened and unscreened, and put their equity curves on the same chart. That picture, more than any paper, will teach you what a confound looks like and why "small caps outperform" was true and useless at the same time.

Then compute your capacity honestly, using a real volume constraint. Most people discover that the version of the strategy that works is the version they are too big to run, or the version they are too small to fund. Knowing which of those you are is the actual decision here.