Quality Minus Junk

Overview

Quality Minus Junk, from Asness, Frazzini and Pedersen, starts from a question that sounds almost too obvious to ask: what should a rational investor be willing to pay more for?

The answer is not controversial. You should pay more for a company that is more profitable, that is growing that profitability, that is safe (low debt, stable earnings, low bankruptcy risk), and that pays out to shareholders. Call all of that "quality".

The empirical finding is that investors do pay more for quality, but not nearly enough. Quality companies do command higher prices, just not high enough to erase their advantage. So a portfolio that goes long high-quality names and short low-quality "junk" names has historically earned a positive return, in the US and in most other developed markets, over a very long sample.

The uncomfortable part, which the authors are open about, is that this means the market persistently underprices something extremely visible. Nobody has a fully satisfying explanation for why.

Thesis (why the edge exists)

Several candidate explanations, none of them airtight.

Leverage constraints. Many investors cannot use leverage. If you want high returns and cannot borrow, you buy risky, high-beta, low-quality stocks instead. That bids up junk and leaves quality relatively cheap. This is the same mechanism behind betting against beta, and the two strategies are cousins.

Lottery preferences. Retail investors like stories, moonshots and stocks that could 10x. Junk provides that. Boring, profitable, safe compounders do not. The demand for lottery tickets suppresses the price of the boring stuff.

Benchmark hugging. Institutional managers are judged against a benchmark over short windows. Being right slowly, which is what quality does, is career-risky. Being wrong quickly alongside everyone else is not.

The honest possibility. Some of the premium may be compensation for a risk we have not identified, or it may be partly a product of researchers trying many definitions of quality until one worked. The strategy is more discretionary than most, which is exactly the environment where data mining thrives. Take that seriously.

Strategy logic

Quality is not one number. It is a composite built in three blocks.

• Profitability. Is the business making money on what it owns? Gross profit to assets, return on equity, return on assets, operating cash flow to assets, gross margin, and low accruals (meaning the earnings are backed by real cash rather than accounting entries).
• Growth. Is the profitability improving? Take the same profitability measures and compute how much they have changed over the past five years. A company that is profitable and getting more so is worth more than one that is profitable and slowly decaying.
• Safety. Will the company still be here in ten years? Low market beta, low leverage, low volatility of earnings, and a low score on a bankruptcy risk model.

Standardise every individual metric into a z-score within its sector. Average the metrics within each block. Then average the three blocks. That single number is the quality score.

Rank the universe on it, buy the top, short the bottom.

Parameters (knobs)

• Which metrics. This is the biggest and most dangerous knob. Every added metric is another chance to overfit. Keep each block to four to six measures and do not tune them on the same data you evaluate on.
• Block weights. Equal weight across profitability, growth and safety is the default. Weighting them differently is where researchers quietly fit the past.
• Growth window. Five years is standard. Shorter windows are noisier and pick up cyclical rebounds rather than structural improvement.
• Neutralisation. Sector-neutral is a minimum. Size and beta neutral is better, since quality naturally tilts large and low-beta.
• Portfolio slice. Deciles for research, quintiles for a real book.
• Rebalance. Monthly or quarterly. Quality moves slowly, so quarterly is usually enough.

Portfolio construction

Rank within sector, then build the legs, then hedge the residual exposures.

Left alone, a quality book carries large unintended bets: long large cap, long low beta, and short value (because quality is expensive). If you want quality alpha rather than a repackaged low-beta fund, you must strip those out explicitly with a risk model.

Weight positions by liquidity, cap single names, and pay attention to the short leg's borrow. The junk end of the market is full of heavily shorted names where the borrow fee can be several percent a year, which can swallow the leg's entire expected return.

Costs, capacity and turnover

Turnover is moderate: fundamentals move slowly, but the composite has many inputs and small changes in several of them can move a name across a decile boundary. Expect roughly 60 to 120 percent per year on a quintile book. Buffer zones help a lot here.

Capacity on the long leg is excellent, because quality companies are typically large and liquid.

Capacity on the short leg is the binding constraint. Junk is small, illiquid, volatile and expensive to borrow. Many published QMJ results are far more attractive than what you can actually implement, and the gap lives almost entirely in the short leg. A long-only quality tilt is often the honest deployable version.

Backtest design checklist

• Point-in-time fundamentals with a proper reporting lag. Non-negotiable.
• Survivorship. The junk leg is precisely where bankruptcies happen. Delete them and you delete the short leg's whole reason for existing.
• Borrow feasibility and cost. Model it, or at minimum report what fraction of the short leg was actually borrowable.
• Overfitting discipline. Pick the metric list before you look at the returns. Then, and this is the hard part, do not change it when the returns disappoint. If you must search, hold out a market or a decade you never touch.
• Correlation report. Show the quality score's correlation with beta, size, value and momentum, over time. If the strategy's return disappears once you neutralise those, you have not found quality, you have found low beta wearing a costume.
• The junk rally. Look specifically at April to September 2009, and at the second half of 2020. Quality gets crushed in those windows and you need to know how badly.

Common failure modes

• The definition trap. With dozens of plausible metrics, some combination will look brilliant on any historical sample. This is the strategy most vulnerable to data snooping in the whole equity factor zoo.
• Junk rallies. When risk appetite returns violently after a crash, the worst companies rally hardest, and the short leg does severe damage in a very short window.
• Quality gets expensive. In flights to safety, quality names get bid to extreme multiples. You end up long crowded, overvalued compounders. 2020 to 2022 is the recent worked example.
• Hidden low-beta bet. A large slice of quality's historical return is just the low-beta effect. Neutralise beta and check what survives.
• Short leg impossibility. The names you most want to short are the ones you cannot borrow, or can only borrow at a fee that erases the trade.

Variants

• Long-only quality tilt. Skip the short leg entirely, overlay quality scores onto a benchmark with a tracking-error budget. This is what most real quality funds do, and it sidesteps the borrow problem.
• Quality at a reasonable price. Combine the quality score with a value score, buying good companies that are not already priced for perfection. This is the single most useful variant and it fixes quality's biggest weakness.
• Safety-only. Use just the safety block. It is simpler, more robust and overlaps heavily with the low-volatility anomaly.
• Profitability-only. Just the profitability block, which is essentially gross profitability and is far less prone to overfitting.
• Quality as a filter. Do not trade it. Use it to screen out junk from a value or momentum book. Often the highest-value application.

Our notes and suggestions

Approach this one with more suspicion than usual. Not because the effect is fake, the evidence across markets and decades is genuinely strong, but because the construction has so many degrees of freedom that it is very easy to fool yourself. Write down your metric list, commit to it, and evaluate once.

The practical recommendation for most people is not to run QMJ as a standalone long-short. It is to use the quality score as a screen inside a value strategy. Cheap and good beats cheap alone, and it beats good alone. That combination is well documented, it is intuitive, and it does not depend on being able to borrow the worst companies on the exchange.