Accruals and Earnings Quality

Overview

A company reports a profit of 100 million. Good news? It depends entirely on where that 100 million came from.

If customers paid cash and the money is in the bank, that is a real profit. If instead the company shipped a lot of product to distributors on credit, booked the revenue, and is now sitting on a mountain of receivables it may or may not ever collect, the reported profit is the same number but the economic reality is completely different.

The gap between reported earnings and actual cash generated is called accruals. Richard Sloan showed in 1996 that this gap predicts future returns. Companies with high accruals, meaning profits that are heavily made of accounting entries rather than cash, systematically underperform over the following year. Companies with low accruals, whose profits are backed by cash, outperform.

The market, apparently, reads the earnings line and does not check the cash flow statement.

Thesis (why the edge exists)

Two mechanisms, and both are real.

Accruals are less persistent than cash. This is the part that is not about behaviour at all, it is about accounting. Cash earnings tend to repeat next year. Accrual earnings tend not to, because they represent timing differences that reverse. A receivable eventually gets collected or written off. Inventory eventually gets sold or marked down. So a company with high accruals today will, mechanically, tend to report worse earnings next year.

Investors do not adjust for this. They see this year's earnings, extrapolate them forward, and price the stock accordingly. They do not discount the accrual portion for its lower persistence. When next year's earnings come in lower than expected, the stock falls. That is the return.

There is also a darker version. High accruals are the natural fingerprint of earnings management: a company under pressure to hit a number can pull revenue forward or push expenses back, and both show up as accruals. So the accrual signal partly picks up companies that are quietly cooking the books, and those tend to end badly.

The clean way to think about it: accruals are a measure of how much of a company's profit is a promise rather than a payment.

Strategy logic

• Compute the gap. The cash-flow version is the cleanest: take net income, subtract cash flow from operations. What remains is total accruals. Divide by average total assets so a large company and a small one are comparable.
• Rank within sector. Working capital cycles vary enormously. A homebuilder carries huge inventory by nature; a software company carries almost none. Comparing them directly is meaningless. Compare a homebuilder to other homebuilders.
• Sort and trade. Buy the lowest-accrual quintile (cash-backed earnings), short or exclude the highest-accrual quintile (promise-backed earnings).
• Rebalance around filings. The signal only updates when the cash flow statement is published, so trade off filing dates rather than a fixed calendar.

Parameters (knobs)

• Accrual definition: the cash-flow-statement version (net income minus operating cash flow) or the older balance-sheet version (change in non-cash working capital minus depreciation). The cash-flow version is more accurate; the balance-sheet version is what the original paper used and what many datasets provide.
• Scaling: average total assets is standard. Sales is an alternative and behaves differently for asset-light firms.
• Sector granularity: sector or industry group. Finer is better here than for most factors, because working capital norms are so industry specific.
• Percent accruals. Divide accruals by the absolute value of earnings instead of by assets. This variant has held up better in later research.
• Rebalance: quarterly, tied to filings.

Portfolio construction

Within-sector z-score, long the lowest, short or exclude the highest.

Because the accruals signal is strongest in smaller companies, an unconstrained book will drift small, and that is where the costs are. Decide deliberately how much small-cap exposure you are willing to take on, and cap it.

The signal also correlates with quality and, weakly, with value. Run those correlations before deploying, or you will discover that your "new" strategy is 70 percent of a strategy you already own.

Diversify. Individual accrual signals are noisy, and any single company can have a perfectly innocent reason for a big working capital swing (a new factory, a seasonal build, an acquisition). The edge is statistical, across hundreds of names, not diagnostic on any one of them.

Costs, capacity and turnover

Turnover is low. Accruals update once a quarter and are reasonably persistent, so a quintile portfolio might turn over 50 to 80 percent a year. That part is friendly.

The costs that hurt are concentration and borrow. The strongest part of the signal historically lived in small caps, and small caps are expensive to trade. If you restrict yourself to large, liquid names, a large chunk of the documented return goes away. That is not a reason to trade illiquid names, it is a reason to be honest about what is left.

On the short side, high-accrual companies are often already crowded shorts, with correspondingly expensive borrow, and they can stay aloft for a very long time before the accounting catches up with them.

Backtest design checklist

• Filing dates, not fiscal dates. Accruals are computed from the cash flow statement, which is published weeks or months after the period ends. Using fiscal-period-end data on the fiscal period end date is a straightforward look-ahead and it makes this strategy look far better than it is.
• Point-in-time financials. Restated cash flow statements are a real thing and they will poison the backtest.
• Financials excluded. A bank's balance sheet does not have working capital in any comparable sense. Leave them out.
• Size decomposition. Split the return by market cap decile. If almost all of it comes from the smallest names, say so out loud, because that changes what the strategy is.
• Sub-period decay. Test 1985 to 1999 and 2005 to today separately. The effect is meaningfully weaker in the later period, which is what you would expect from a widely published anomaly.
• Merger and acquisition noise. An acquisition creates a huge apparent working capital change that has nothing to do with earnings quality. Flag or exclude acquisitive companies.

Common failure modes

• Decay. This is the headline risk. Published in 1996, widely taught, embedded in every commercial risk model. The premium is a fraction of what it was, and honest backtests show that clearly.
• Look-ahead through filing dates. The single most common bug, and it flatters the strategy enormously.
• Innocent explanations. A company building a new plant or ramping inventory ahead of a genuine demand surge will look like a high-accrual company. The signal cannot distinguish good growth from bad accounting.
• Short leg pain. Aggressive-accounting companies can keep the story running for years, and shorting a stock with a good narrative is a way to lose money while being right eventually.
• Small-cap dependence. If the return only exists in names you cannot trade, it does not exist.

Variants

• Percent accruals. Scale by earnings instead of assets. Has held up better out of sample than the original.
• Cash-based operating profitability. Combine profitability and accruals into one measure: profit that is actually backed by cash. Ball, Gerakos, Linnainmaa and Nikolaev showed this absorbs much of the accruals effect and is a cleaner signal.
• Net operating assets. The balance-sheet cousin of accruals, measuring the cumulative gap between accounting and cash over the company's whole history.
• Accruals as a screen. Rather than trading it standalone, use it to filter a value or quality book. Screening out the highest-accrual names from a value portfolio removes a lot of the value traps.
• Combine with short interest. High accruals plus rising short interest plus insider selling is a much sharper red flag than accruals alone.

Our notes and suggestions

The most valuable thing about this strategy is not the strategy, it is the habit it forces. Once you have built an accruals signal, you can never again look at an earnings number without asking whether the cash showed up. That question is the foundation of every serious piece of fundamental analysis, systematic or otherwise.

As a standalone trade in liquid large caps today, be sceptical. The effect is old, famous and much diminished. As a screen inside a value or quality strategy, it still earns its place, cheaply, and it removes a category of disaster that no price-based signal will ever see coming.

Build it, run the pre-2000 and post-2005 backtests side by side, and take the decay seriously. It is one of the clearest, most honest illustrations in all of quant finance of what happens to an edge after everybody knows about it.