Semi-deviation leaves the risk ranking of these eight Central and Eastern European markets almost untouched. Poland and Czechia are the sole CEE pair to switch places, separated by standard deviations of 5.05% and 5.04%. Include the euro area benchmark row from Table 3, and Slovenia also changes places with the benchmark. This sample gives little support to the claim that semi-deviation tells a trader something total risk misses.
The study Hatipoğlu ran
Hatipoğlu audits eight national equity indices: Bulgaria, Croatia, Czechia, Hungary, Poland, Romania, the Slovak Republic and Slovenia. A euro area index provides the benchmark. The data are monthly OECD share price indices, 2015 = 100, from January 2002 to March 2025, converted to first log differences. Each series has roughly 279 monthly observations, falling to 278 after differencing.
The premise will be familiar to traders. Returns are not normal, and the standard deviation treats a 25% up month like a 5% down month even though investors experience them differently. Hatipoğlu documents the departure from normality: kurtosis exceeds 3 for all nine series, while Jarque-Bera p-values are 0.00 throughout. Every market is negatively skewed except the Slovak Republic at +0.79. Bulgaria supplies the extreme observation, with skew of -1.87, kurtosis of 13.54 and a worst month of -43.23%.
Four tools carry the analysis. Semi-standard deviation is measured relative to the sample mean. Value at Risk is estimated at 95% and 99%. A single-factor regression places country excess returns against euro area excess returns, with the long-term German government bond yield as the risk-free rate, and produces beta plus a residual standard deviation labelled idiosyncratic risk. Finally, a GARCH(1,1)-in-Mean model with Student-t errors puts conditional risk into the mean equation and interprets its coefficient delta as a risk premium.
The main figures fit in a few lines. Mean monthly returns range from 1.13% for Romania to 0.32% for the Slovak Republic, compared with 0.21% for the euro area. Standard deviations extend from 3.87% in the Slovak Republic to 6.40% in Bulgaria. All betas are below one, spanning 0.27 to 0.85. Delta is positive and significant everywhere except the Slovak Republic.
Does downside risk alter the order?
All eight markets have lower semi-deviation than total volatility. Bulgaria records 4.98% against 6.40%, Czechia 4.00% against 5.04%, and the Slovak Republic 2.50% against 3.87%. The paper presents this difference as a result and bases its conclusion on it: models that calculate downside risks, it argues, permit more detailed measurement of CEE stock exchange performance than models based on total risk.
The gap follows directly from the formula. It sums squared deviations only for months below the target, then divides by the same T-1. The downside sum is therefore a subset of the total sum, so its result cannot exceed the standard deviation. The calculation reveals nothing particular about CEE markets.
Ranking is the informative test, and the paper does not perform it. Standard deviation orders the series as Bulgaria, Romania, Croatia, Hungary, Poland, Czechia, Slovenia, euro area, Slovak Republic. Semi-deviation gives Bulgaria, Romania, Croatia, Hungary, Czechia, Poland, euro area, Slovenia, Slovak Republic. Poland and Czechia switch places because total volatility differs by 5.05% versus 5.04%. Slovenia trades places with the euro area benchmark.
Seven of the eight CEE markets retain the same rank.
Calling those two columns a more detailed measurement is hard to sustain.
The 99% VaR column behaves differently. Both volatility measures identify the Slovak Republic as the safest market. Its 95% VaR, at 6.83%, ranks second lowest among nine, behind only the euro area at 5.85%. At 99%, the Slovak Republic falls to fifth of nine with 14.80%, higher than Poland at 13.29% and Czechia at 11.98%. The paper confines its low-loss claim for the Slovak Republic to the 95% level and does not report its relative position at 99%.
A Slovak figure that cannot be historical
The Slovak 14.80% exceeds the worst monthly return observed in the sample, which Table 2 reports as -14.66%. With 279 observations, historical VaR at 99% lies around the third-worst month and cannot exceed the minimum. The estimator must therefore be distributional. We did not find its specification anywhere in the paper. The method section merely describes VaR as suitable for all types of return distributions. We also found no exception-rate backtest.
This omission bears directly on the paper's premise. Non-normality motivates the exercise, yet the unstated tail distribution determines the full 99% column.
The Slovak Republic looks unusual under every measure. It alone has positive skew, at +0.79, and it has the lowest beta, at 0.27. Its residual standard deviation is 3.94%, above its own return standard deviation of 3.87%, meaning the euro area factor explains essentially none of the variance. Its GARCH-M delta is the only insignificant estimate, although its point estimate of 0.62 is the table's largest. These findings appear in separate sections and are never connected. The paper comes closest in one sentence: "in the case of the Slovak Republic stock exchange, returns should be explained by factors other than volatility."
Bulgaria offers a useful comparison. Residual risk of 5.67% against total volatility of 6.40% suggests that the factor captures roughly a fifth of the variance. For Hungary, 3.15% against 5.48% suggests around two thirds. These approximations compare residuals from excess returns with volatilities calculated from raw returns.
What does delta buy you?
Delta is positive and significant in eight of nine series. Estimates range from 0.12 in Croatia to 0.26 in Hungary, while the euro area stands at 0.20. Persistence remains below one throughout, from 0.84 to 0.95. ARCH LM(1) p-values between 0.11 and 0.97 indicate a clean variance equation. Bulgaria's ARCH term of 0.40, compared with Poland's 0.08 and Hungary's 0.10, shows genuine heterogeneity.
Its meaning in return units cannot be recovered. Equation 4 specifies the in-mean term as conditional variance. Yet most conditional series plotted in the appendix remain between 0.00 and 0.12, while Romania, Croatia and Poland reach 0.20 in places. Those values resemble standard deviations for markets whose monthly volatility runs from 3.87% to 6.40%. Interpreted as variances, they would be absurd.
The choice changes the result materially. Reading the term as variance, Hungary's 0.26 contributes about 0.08% a month at average variance. Reading it as standard deviation raises the contribution to about 1.4% a month, above Hungary's entire realised mean of 0.88%. Those interpretations amount to different papers. Across nine separately estimated in-sample specifications, the tables show significance stars without standard errors or t-statistics and apply no multiple-testing adjustment. A reader therefore cannot gauge the uncertainty.
Full-sample estimation also limits the betas. They cover the entire 2002 to 2025 window even though the paper identifies volatility spikes in 2008-09 and 2020, followed by a secular decline since the early 2000s. We did not find rolling or subperiod estimates. We raised the same concern in a GARCH study of the Nepalese market, where full-sample persistence fit neither side of a mid-sample break.
The OECD series are price indices, so dividends are excluded. Currency treatment poses the larger problem. Only Slovenia, the Slovak Republic and, from 2023, Croatia spend any substantial part of the sample inside the euro. We found no currency conversion described for the regression against the euro benchmark.
Several discrepancies also weaken confidence in the tables. The narrative reports Slovak downside risk as 1.40%, whereas Table 3 gives 2.50%. The text says theta and beta are significant except for Hungary, though the table marks Hungary's theta at 10% and its beta at 1%. It also names Romania as having the highest potential loss at both confidence levels. Table 3 instead puts Bulgaria higher at 95%, with 10.53% against Romania's 10.34%.
We could not test any of these results ourselves. Our data does not include the monthly OECD share price indices for the eight markets or the euro area benchmark used in the regressions. Substituting US index series would answer a different question.
The claim remains too large
The region merits study. Table 1 covers seven of the eight markets, with Hungary absent and blank market-cap cells for the Slovak Republic and the euro area. Together they account for about 1.60% of world GDP in 2022 against 0.27% of world market capitalisation. The VaR estimates imply maximum monthly losses of roughly 10% to 20%.
A ranking in which semi-deviation and standard deviation differed by more than the 0.01 percentage point Poland/Czechia gap, or reordered the top three, would change my view of the downside-measure claim. Here the rankings agree.