How to Calculate Standardized Unexpected Earnings (SUE)

Introduction

In 1989, Victor Bernard and Jacob Thomas published a paper in the Journal of Accounting and Economics that documented one of the most robust anomalies in finance: post-earnings announcement drift (PEAD). Stocks with positive earnings surprises continued to drift upward for weeks after the announcement, and stocks with negative surprises continued to drift downward. The market, in other words, did not fully react to earnings news at the time of announcement.

The metric Bernard and Thomas used to measure the magnitude of the earnings surprise was Standardized Unexpected Earnings (SUE). By sorting stocks into deciles based on their SUE scores, they demonstrated that a long-short portfolio — long the top decile, short the bottom decile — generated abnormal returns of approximately 4% over the 60 trading days following the earnings announcement.

Decades later, PEAD remains one of the most well-documented anomalies in academic finance. And SUE remains the standard metric for measuring it. This article explains the formula, walks through the calculation step by step, examines the academic history, and shows how to implement SUE in practice.

The SUE Formula

The core idea behind SUE is simple: an earnings surprise should be measured not just by its magnitude, but relative to the typical size of surprises for that company. A $0.05 earnings beat for a company that routinely beats by $0.04 is not very surprising. A $0.05 beat for a company that usually misses by $0.02 is very surprising. SUE captures this by normalizing the surprise by its historical standard deviation.

Each component of the formula serves a specific purpose:

• Actual EPS: The earnings per share reported by the company for the quarter.
• Expected EPS: The earnings per share that the market expected. This can be estimated using either analyst consensus forecasts or a time-series model (more on this below).
• Standard deviation of past surprises: The historical variability of the company's earnings surprises, computed over a trailing window of prior quarters. This denominator is what makes the metric "standardized" — it normalizes across companies with different levels of earnings predictability.

Academic History

The concept of standardized unexpected earnings did not emerge fully formed. It was developed and refined over two decades of accounting and finance research:

Ball and Brown (1968)

The story begins with Ray Ball and Philip Brown's landmark 1968 paper in the Journal of Accounting Research. Ball and Brown documented that stocks with positive earnings surprises experienced positive abnormal returns around the announcement date, and stocks with negative surprises experienced negative abnormal returns. This was the first systematic evidence that earnings announcements contain information that moves stock prices — and that some of the price adjustment happens after the announcement, not instantaneously.

Latané and Jones (1977)

Henry Latané and Charles Jones were the first to propose standardizing the earnings surprise. In their 1977 paper in The Financial Analysts Journal, they argued that raw earnings surprises are not comparable across companies. A $0.10 surprise means very different things for a company with volatile earnings versus a company with stable earnings. By dividing the surprise by its historical standard deviation, the metric becomes a z-score that is comparable across the entire cross-section of stocks. This was the birth of SUE as a standardized, cross-sectionally comparable metric.

Foster, Olsen, and Shevlin (1984)

George Foster, Chris Olsen, and Terry Shevlin published a comprehensive study in The Accounting Review in 1984 that refined the SUE methodology and documented the drift pattern across a large sample. They used a time-series model to estimate expected earnings (a seasonal random walk with drift) and showed that the drift persisted for at least 60 days after the announcement. Their paper established the methodological foundation that Bernard and Thomas would build upon.

Bernard and Thomas (1989)

Victor Bernard and Jacob Thomas's 1989 paper in the Journal of Accounting and Economics is the definitive study on PEAD. Using SUE deciles constructed from a seasonal random walk model, they documented that the long-short spread between the top and bottom SUE deciles generated approximately 4% abnormal returns over 60 trading days post-announcement. They also provided evidence on the mechanism: much of the drift was attributable to the market's failure to fully incorporate the implications of current earnings for future earnings (specifically, the autocorrelation in quarterly earnings surprises).

Bernard and Thomas showed that the drift was not explained by risk, transaction costs, or methodological artifacts. It appeared to be a genuine pricing inefficiency driven by investors' under-reaction to earnings news.

Two Approaches to Expected EPS

The numerator of the SUE formula requires an estimate of "expected" EPS. There are two standard approaches, and the choice has practical implications:

Approach 1: Analyst Consensus Forecast

The most common approach in practice is to use the median analyst estimate from services like I/B/E/S (now part of LSEG/Refinitiv), Bloomberg consensus, or FactSet. This represents the market's collective expectation for the company's earnings.

The advantage of using analyst consensus is that it reflects the information actually available to the market. If the consensus expects $1.50 and the company reports $1.60, the surprise is $0.10 — and this genuinely represents new information relative to the market's prior belief.

The disadvantage is that analyst forecasts can be biased. Research has documented persistent optimism bias in analyst estimates, particularly at longer forecast horizons. Analysts may also herd toward consensus, reducing the dispersion of forecasts. Additionally, analyst coverage is not universal — small-cap and micro-cap companies may have few or no analyst estimates, making this approach unavailable.

Approach 2: Time-Series Model (Seasonal Random Walk)

The approach used by Bernard and Thomas in their original work was the seasonal random walk with drift. Under this model, the expected EPS for quarter q is equal to the EPS from the same quarter one year ago (q - 4) plus a drift term:

This model captures the seasonality inherent in quarterly earnings (Q4 for a retailer will always look different from Q2) while accounting for a secular trend. The surprise is then: Actual EPS_q - (EPS_q-4 + drift).

The advantage of the time-series approach is that it requires only the company's own earnings history — no analyst coverage needed. This makes it applicable to the full universe of public companies. The disadvantage is that it ignores information the market has that goes beyond the company's own earnings trend, such as changes in competitive position, macro conditions, or management guidance.

Which Approach Is Better?

Research suggests that both approaches produce PEAD signals, but the analyst consensus approach tends to produce stronger drift in modern data. This makes intuitive sense: the analyst consensus captures what the market actually expects, so deviations from it are genuine surprises to the marginal investor. The time-series model, by contrast, may flag "surprises" that the market had already anticipated based on other information.

In practice, many implementations use analyst consensus as the primary approach and fall back to the time-series model for companies without analyst coverage.

Step-by-Step Calculation

Let us walk through a complete SUE calculation with concrete numbers.

Step 1: Gather Earnings History

Suppose we are calculating SUE for Company XYZ's Q1 2026 earnings. We need the company's actual EPS and the analyst consensus estimate for Q1 2026, plus the earnings surprises for the prior 8 quarters:

Step 2: Calculate the Standard Deviation of Past Surprises

The 8 historical surprises are: +0.05, -0.05, +0.03, +0.05, -0.02, +0.04, +0.03, +0.03.

Mean surprise = (0.05 - 0.05 + 0.03 + 0.05 - 0.02 + 0.04 + 0.03 + 0.03) / 8 = 0.16 / 8 = 0.02

Variance = [(0.05-0.02)² + (-0.05-0.02)² + (0.03-0.02)² + (0.05-0.02)² + (-0.02-0.02)² + (0.04-0.02)² + (0.03-0.02)² + (0.03-0.02)²] / 7

= [0.0009 + 0.0049 + 0.0001 + 0.0009 + 0.0016 + 0.0004 + 0.0001 + 0.0001] / 7

= 0.0090 / 7 = 0.001286

Standard deviation = sqrt(0.001286) = 0.0359

Step 3: Compute SUE for the Current Quarter

Now suppose Company XYZ reports Q1 2026 actual EPS of $1.42 against a consensus estimate of $1.35:

A SUE of +1.95 means the earnings surprise was approximately 1.95 standard deviations above the company's typical surprise magnitude. This is a genuinely surprising beat — substantially larger than what the market has historically seen from this company. In Bernard and Thomas's framework, this stock would be placed in one of the top SUE deciles and would be expected to experience positive drift over the following weeks.

Why SUE Matters More Than Simple EPS Beat/Miss

Many financial news outlets report earnings as a simple binary: "beat" or "miss." But this binary classification destroys information. SUE preserves the magnitude and context of the surprise, which is critical for several reasons:

The Same Dollar Beat Can Mean Different Things

Consider two companies that both beat consensus by $0.03 in the same quarter:

• Company A has historically had surprises with a standard deviation of $0.02. SUE = $0.03 / $0.02 = +1.50. This is a notable beat — 1.5 sigma above normal.
• Company B has historically had surprises with a standard deviation of $0.15. SUE = $0.03 / $0.15 = +0.20. This is within the noise — only 0.2 sigma. The market will barely react.

The simple "beat by $0.03" headline treats these two situations identically. SUE reveals that Company A's beat is genuinely informative, while Company B's beat is essentially meaningless relative to its historical variability.

Persistent Beaters Are Correctly Penalized

Some companies beat analyst estimates almost every quarter. This is well-documented — the "expectations management" game where companies guide analysts down and then reliably beat the lowered bar. For these companies, a $0.02 beat is not surprising at all. SUE handles this correctly: if the company has a history of similar beats, the historical standard deviation will be small relative to the raw surprise, but the mean surprise will also be positive. The SUE score for a "typical" beat will be moderate, not extreme. Only a beat that is larger than the company's own historical pattern will generate a high SUE.

Misses Matter Too

SUE works symmetrically. A company that typically beats by $0.03 but this quarter misses by $0.01 will have a negative SUE — the surprise is meaningfully below the company's own trend, even though the raw miss is small. The standardization captures the change in pattern, which is often more informative than the absolute magnitude.

Implementing SUE in Practice

For practitioners who want to calculate SUE systematically, there are several implementation decisions to make:

Data Sources

The two critical inputs are actual EPS and expected EPS. For analyst consensus estimates, the most widely used academic source is I/B/E/S (Institutional Brokers' Estimate System), now maintained by LSEG (formerly Refinitiv). Bloomberg and FactSet provide comparable consensus data. For free or low-cost alternatives, many financial data APIs provide consensus estimates, though the data may be less comprehensive for small-cap stocks.

For actual EPS, the company's earnings release and SEC filings (10-Q and 10-K) are the authoritative sources. Many data providers also offer actual EPS data that is already matched to the consensus estimate on a comparable basis (same accounting adjustments, same share count).

Lookback Window

The standard choice for the denominator is the trailing 8 quarters (2 years) of earnings surprises. This provides enough data to compute a meaningful standard deviation while being recent enough to reflect the company's current earnings characteristics. Some implementations use 12 or 16 quarters; the tradeoff is between statistical stability (more quarters) and relevance (fewer quarters, more recent).

Handling Edge Cases

Several practical edge cases arise in SUE calculation:

• Zero or near-zero standard deviation: If a company has virtually identical surprises every quarter (e.g., always beats by exactly $0.02), the standard deviation approaches zero, and SUE can blow up to extreme values. A common solution is to impose a minimum floor on the denominator — for example, the greater of the calculated standard deviation or $0.01.
• Insufficient history: Newly public companies or companies with fewer than 8 quarters of data require special handling. Some implementations use whatever history is available (minimum 4 quarters); others exclude the company until sufficient data accumulates.
• Restatements and one-time items: If a prior quarter's EPS was restated, the surprise for that quarter should be recalculated using the original (pre-restatement) consensus and the restated actual. One-time items (restructuring charges, legal settlements) can distort both the surprise and the historical standard deviation. Many implementations use "adjusted" or "operating" EPS to filter out these effects.
• Share count changes: Stock splits, buybacks, and dilution can affect EPS comparisons across quarters. Using per-share data on a consistent, split-adjusted basis is essential.

A Python Implementation Sketch

For investors working in Python, the yfinance library provides earnings date and estimate data that can be used as a starting point for SUE calculation:

The ddof=1 parameter uses the sample standard deviation (Bessel's correction), which is appropriate when estimating the population standard deviation from a sample. The 0.01 floor prevents extreme SUE values for companies with very stable earnings.

To compute SUE at scale, you would loop through each company in your universe, retrieve its earnings history, calculate the trailing 8-quarter surprises, and apply the formula. The resulting SUE scores can then be ranked across the entire cross-section to identify the highest and lowest deciles.

From SUE to PEAD: Building a Trading Strategy

The standard PEAD strategy, as documented by Bernard and Thomas, works as follows:

1. At each earnings announcement date, compute the SUE for every company that reported.
2. Sort companies into deciles based on their SUE scores across the entire cross-section of that quarter's reporters.
3. Go long the top SUE decile (largest positive surprises) and go short the bottom SUE decile (largest negative surprises).
4. Hold the portfolio for 60 trading days (approximately one calendar quarter).
5. Repeat each quarter as new earnings are announced.

Bernard and Thomas found that this strategy generated approximately 4% abnormal returns over the 60-day holding period. Subsequent research has confirmed the drift, though its magnitude has varied across time periods and may have diminished somewhat in recent decades as more capital has been allocated to exploit it.

SUE and the Gap-Volume Proxy

Not every data source provides clean analyst consensus estimates for every company. For companies without analyst coverage, or when using free data sources, a common fallback is the gap-volume proxy: the stock's percentage gap at the open on the earnings announcement date, weighted by the ratio of announcement-day volume to the trailing average volume.

The intuition is straightforward: a large gap on high volume after an earnings announcement is the market's revealed assessment of the surprise magnitude. If the stock gaps up 5% on 3x normal volume, the market is telling you something significant and unexpected happened. This is not SUE — it does not use EPS data at all — but it captures a similar concept: the market's reaction, normalized for the stock's typical behavior.

Alpha Suite's PEAD scanner uses the analyst-consensus-based SUE as the primary metric where estimates are available, and falls back to the gap-volume proxy for companies without analyst coverage. This hybrid approach ensures broad coverage across the entire US equity universe, including small-cap and micro-cap stocks that are often the most fertile ground for PEAD signals (due to lower analyst attention and greater information asymmetry).

SUE vs. Other Earnings Surprise Metrics

SUE is not the only way to measure earnings surprise. Other approaches include:

• Raw surprise (Actual - Estimate): Simple but not cross-sectionally comparable. A $0.05 surprise means different things for different companies.
• Percentage surprise ((Actual - Estimate) / |Estimate|): Better than raw surprise, but can produce extreme values when the consensus estimate is near zero.
• Earnings surprise scaled by price (surprise / stock price): Normalizes by share price, making it somewhat comparable across companies. Used in some academic studies.
• Consensus surprise decile rank: Ranks the percentage surprise within the cross-section of that quarter's reports, producing a percentile. Simple and robust but discards information about the company's own historical surprise pattern.

SUE's advantage is that it normalizes by the company's own historical variability, capturing both the cross-sectional and time-series dimensions of "surprise." A high SUE means the surprise is large relative to what this specific company typically delivers — which is a stronger predictor of drift than a surprise that is large in absolute terms but typical for that company.

Conclusion

Standardized Unexpected Earnings is a deceptively simple metric with deep roots in academic finance. First formalized by Latané and Jones in 1977, refined by Foster, Olsen, and Shevlin in 1984, and made famous by Bernard and Thomas's 1989 PEAD paper, SUE remains the standard measure of earnings surprise magnitude. Its power comes from a single insight: an earnings surprise should be measured relative to the company's own historical pattern of surprises, not in absolute terms.

For quantitative investors, SUE is a building block. It can be used standalone to construct PEAD strategies, or combined with other signals — insider buying conviction, technical momentum, volatility regime — to build multi-factor models. Understanding how to calculate it correctly, including the choice of expected earnings model and the handling of edge cases, is foundational knowledge for anyone working with earnings data.