Colombo's value-weighted operating-profitability spread earned 0.78% a month, enough to matter economically and far too uncertain to trade. The book bought firms with the highest operating profitability and sold those with the lowest from April 2010 to December 2023 (t 1.0643, p 0.2887). Twelve months of that return amount to about 9% a year before trading costs. The authors describe the premium as "weak and tentative," conceding a small premium rather than none. They also write that "these effects are economically small and statistically insignificant." The statistical judgement holds. Their own value-weighted estimate contradicts the word "economically".

What the portfolios hold

Jayasinghe, Fernando and Pathirawasam study listed non-financial firms on Sri Lanka's Colombo Stock Exchange (CSE). They exclude firms with missing accounts or extreme outliers, then construct two accounting signals. Operating profitability (OP) divides operating income by total assets; cash profitability (CP) divides cash flow from operations by total assets. The measures follow Fama and French (2015) and Novy-Marx (2013).

Each month, the authors rank firms separately on each signal and group them from least to most profitable. They call the return difference between the top and bottom groups RMW. Fama and French (2015) supply the economic argument: profitable firms carry either lower risk or better investment opportunities, and either should command a premium. The expected spread is therefore positive.

The authors also run monthly cross-sectional regressions on beta, size, value, investment and momentum. They add variables sequentially, starting with beta and ending with momentum, and repeat the regressions within each sort group. Their verdict is that "neither operating nor cash profitability consistently predicts stock returns once these factors are considered."

Full-sample OP earns 0.20% a month equal-weighted (t 0.5880), against 0.78% value-weighted (t 1.0643). CP returns are -0.11% (t -0.1219) and -0.22% (t -0.2142), respectively. No regression characteristic approaches a t of 2. Despite monthly rebalancing, we found no transaction-cost adjustment anywhere.

Is 0.78% a month economically small?

About 9% a year is economically substantial. The authors have grounds to call the estimate statistically weak; the point estimate does not support calling it economically small.

A return of 0.0078 with a t of 1.0643 implies a monthly standard error near 0.73%. The resulting 95% band extends from about -0.7% to +2.2% a month. It accommodates no premium at all, as well as a premium above 2% a month. The value-weighted OP estimate is too noisy to settle the question.

The tables also pull against the paper's account of weighting. The authors say equal-weighted portfolios, "which emphasize smaller firms, exhibit somewhat stronger RMW effects," citing Novy-Marx on size dependence. Their conclusion is more guarded: "Equal-weighted strategies may capture slightly stronger effects due to greater exposure to smaller firms." If exposure to small firms drove the spread, equal weighting should deliver the larger return. Yet value weighting produces the larger OP spread in four of the five cuts:

Equal weighting wins only in 2010 to 2019, when both spreads are negative (-0.0066 against -0.0108). The strongest defence of the authors' wording is the equal-weighted up-market spread: it has the table's highest t-statistic. That ranks precision. By magnitude, the value-weighted spread wins.

Cash loses; operating profitability follows the market

Both full-sample CP spreads are negative. The authors acknowledge that CP "fails to distinguish higher-returning firms." The interesting cell is in OP during up markets, where equal-weighted returns rise from 3.89% in the bottom group to 4.95% in the top group. That gives a spread of 1.06%, with p 0.0666. During down markets, the same spread falls to -0.56% (t -1.4513). The authors interpret profitability as "a procyclical characteristic rather than a defensive quality factor."

Tables 1 and 2 contain 20 RMW spreads across two signals, two weightings and five cuts: the full sample, up markets, down markets, 2010 to 2019, and 2019 to 2023. Of those 20 p-values, one falls below 0.10. Chance alone would yield about two. We also could not find a definition of up and down months. If the authors classify regimes using the same month's market return, a trader cannot trade the conditional spread in advance.

How many cash-profitability groups?

The abstract and methodology describe CP quartiles. Section 3.2 instead specifies quintiles "for each proxy". Tables 2 and 5 contain five CP groups. Table 5's accompanying text refers to "each cash profitability quintile" and says "Portfolios are sorted into five groups based on CP," leaving the abstract and methodology alone in describing quartiles. One description must be wrong. The paper does not identify which.

We found no monthly firm counts either. In a small market, a fifth of the universe could contain only a handful of names.

The return series also warrants checking. In Table 1, every equal-weighted OP portfolio averages between -0.50% and -0.94% a month over the full sample. Value-weighted CP group 4 in Table 2 averages -1.64% over that same period. Compounding -0.94% a month for roughly 165 months leaves about a fifth of the starting capital. Before trusting these returns, a reader would want them reconciled with the CSE index.

Averaging the t-statistics

Tables 4 and 5 place "Average T-Statistics" beside average coefficients. Every reported value is below 0.4 in absolute value. In the standard Fama-MacBeth test, the mean monthly slope is divided by its time-series standard deviation over the square root of T. Taking the average of each month's own t answers another question.

A small slope that stays stable across months can have a large Fama-MacBeth t alongside tiny monthly t's. With about 165 months, the difference can approach an order of magnitude. The methodology mentions heteroskedasticity-robust standard errors, consistent with per-month t's. The paper never specifies which statistic appears in Table 3.

Table 3 presents another problem. The text reports investment slopes from -0.0002 to -0.0003. In the table's longer rows, the second-to-last entries run from -0.000001 to -0.000011. Those would be investment if the columns read from left to right. The layout itself makes matching coefficients to variables difficult.

The abstract offers profitability as a fallback, saying it "should therefore be viewed as a complementary rather than a standalone determinant." Table 3 includes profitability alongside value and momentum in a regression, but no t-statistic there exceeds 0.62. The other characteristics do no better. Momentum has a slope of -0.224673 (t -0.367117) in one multivariate specification; beta on its own has a slope of 0.005958 (t 0.589326). The conclusion also recommends macroeconomic indicators, although none enters a regression.

After 2019, the OP spread changes sign

From 2010 to 2019, OP spreads are -0.0066 equal-weighted and -0.0108 value-weighted. From 2019 to 2023, they become +0.0128 and +0.0236. None of the four reaches significance. The subperiod labels also overlap in 2019. The evidence covers one frontier exchange over nearly fourteen years (165 months). Even the 2019-2023 value-weighted spread of 0.0236 has a t of only 1.0728, implying a standard error near 2.2% a month.

We could not reproduce the study ourselves.

A rerun requires CSE prices and market caps, together with Sri Lankan accounts and their filing dates. We have none of those data. Substituting a US rerun would answer a question about US stocks.

A firm count for each group and a t-statistic explicitly labelled Fama-MacBeth on the mean OP slope would move us. If the value-weighted spread held near 0.78% with a t above 2, Colombo would have a profitability premium worth trading.