Automated analysis

This analysis was drafted by our research engine and has not been checked by a human editor. It may contain errors. It separates the paper’s own results from our tests, and any figures called ours come from our own backtest.

Our automated analysis

Drought changes sign across JSE property indices

A GARCH-MIDAS study finds climate effects, while short-run persistence presses against its limit.

2026-10-15 · 5 min read · Climate-informed volatility forecasting and portfolio risk · South African listed property index exposure (unavailable on the platform)

Reviewing: The Impact of Disaggregated Physical Climate Risk on the Volatility of Property Indices: Evidence from South Africa · Siseko Mtunzi Merana, Hilary Tinotenda Muguto and Lorraine Muguto · Read it on openalex

A risk desk cannot give drought the same volatility sign across five South African property indices. In the coefficient table, drought (DROU) raises long-run volatility for J253 and lowers it for J800. That split matters more than the paper's claim that robustness checks keep any one segment from driving the results, because those checks are not reported. I would use the estimates as hypotheses to test. The short-run model sits at its persistence boundary, and nothing is evaluated out-of-sample.

What Merana and the Mugutos estimated

The paper studies physical climate risk in listed property: drought, heavy rain and temperature extremes, rather than carbon-policy risk. Buildings stay put, so local climate stress can reach their cash flows. Merana, Muguto and Muguto set out the channels. Drought adds water and municipal costs. Heavy rain brings damage and insurance claims. Temperature extremes affect energy use and building operations. They argue that South Africa's infrastructure constraints amplify each channel. Their case for long-run volatility rests on markets absorbing slow-moving physical risk gradually, rather than repricing it all at once.

Daily Bloomberg prices cover five JSE indices: J800 Tradable Property, J803 All-Property, J805 SA REIT, J253 SA Listed Property and J254 Capped Property. The stated window runs from 30 June 2004 to 31 October 2025. The climate inputs come from the Actuarial Society of South Africa (ASSA) Climate Index. Each of its four monthly components has 255 observations: drought (DROU), extreme precipitation (EXPT), extreme minimum (EXMI) and maximum (EXMA) temperature.

GARCH-MIDAS (mixed-data sampling) separates daily variance into a GARCH(1,1) short-run part and a slower long-run part. The log of the long-run part equals a constant plus theta times a weighted sum of the past K months of one climate series. A beta polynomial supplies the weights. BIC and log-likelihood select K from 12, 24 or 36. Five indices crossed with four indicators produce 20 separate fits; inference rests on theta.

The authors say temperature extremes "exert the greatest influence." EXMI theta is positive and significant in all five indices: 11.4460 (J253), 2.8198 (J254) and 2.0198 (J805) at 1%, plus 0.0001 for J800 at 1% (indistinguishable from zero in size) and 0.3575 for J803 at 10%. The EXMA evidence is thinner. Two of five fits are significant, J253 at 2.2977 (1%) and J800 at 1.9208 (5%). J254 (0.3427), J803 (1.6864) and J805 (1.8636) are insignificant. Half the temperature case, then, reaches only two indices out of five.

Which way does drought run?

The abstract says drought "consistently increases long-run volatility, particularly in segmented indices, while broader indices show weaker effects." The authors do flag weaker effects in the broader indices. The five drought thetas give a sharper picture: +4.3505 for J253 and +1.0730 for J254, both at 1%; -5.4098 for J800, significant at 10%; and insignificant estimates for J803 (-4.2629) and J805 (1.1529). Two positive, one negative, two indistinguishable from zero.

For J800, the authors suggest "broader market diversification may partially offset drought-related risk." A partial offset would move a coefficient toward zero. "Weaker" and "partially offset" imply attenuation, while J800's -5.41 reverses sign and exceeds J253's +4.35 in magnitude. A risk manager relying on "consistently increases" would assign drought a sign found in two of five fits.

EXPT also changes direction. It lowers volatility in J253 (-16.9300) and J805 (-1.3957), while raising it in J803 (0.9014, 10%). The authors attribute negative signs to drought relief and positive signs to flood damage. Each indicator is one monthly series, unmatched to the locations of an index's buildings. On that basis, either explanation can be made to fit any sign.

Persistence reaches the limit

The paper imposes alpha + beta < 1, yet alpha + beta falls between 0.9999 and 1.06 in every one of the 20 fits. EXPT on J803 is at the top, with 0.4773 + 0.5827. At one, the short-run component never mean-reverts. Persistence already absorbed by the GARCH part cannot also be credited to climate. The authors interpret significant short-run parameters as strong volatility persistence, consistent with rents and valuations adjusting slowly.

The parameter shape warrants attention as well. Alphas range from 0.3465 to 0.7249, with betas as low as 0.2751. Daily equity data more typically have a small alpha and beta near 0.9. Under drought, J800, J803 and J805 all show alpha 0.4169 and beta 0.5831 to four decimals; J803 and J805 show the same pair again under EXMI. I would want an explanation for three return series landing on identical parameters before using theta from those fits.

Return histories diverge

The observation counts fall short of the stated window for some indices. J253 has 5,539 daily returns and J254 has 5,129, while J800, J803 and J805 each have 2,065. Those 2,065 trading days cover about eight years of the stated 21. The climate series spans the full period at 255 months.

Lag selection mostly favours 36 months. For the three short series, a 36-month window takes up a large share of about eight years (2,065 trading days) of returns, leaving few independent monthly swings to identify theta. The likelihood changes deserve a check too. Under EXMI for J253, log-likelihood moves from -16,497.93 at 12 lags to -8,990.78 at 36; BIC falls from 6.2554 to 3.7875. A gain of about 7,500 log points from reweighting one monthly regressor makes me want to know whether the effective sample changed with lag length.

There are smaller discrepancies. The text reports J253 EXMI theta as 11.05; the table gives 11.4460. EXPT's ADF statistic appears as +14.6606 and is marked significant at 1%. For J253 EXPT, the mean parameter is 0.7183, against a descriptive daily mean of 0.0005.

Fitted coefficients, no forecast

The conclusion says climate risk is "transmitted primarily through volatility" and investors focused on returns "may underestimate exposure." The mean equation contains only a constant.

The price channel is never estimated.

Both estimation tables, for lag selection and coefficients, report in-sample fits. I found no forecast loss against a plain GARCH(1,1), nor a translation of theta into annualised volatility points. The paper cites Wu et al. (2024), where temperature anomalies improve forecasts of aggregate South African stock volatility. A comparable forecast test is what this paper needs. The authors acknowledge that seasonality in the climate data is not modelled; they leave firm-level exposure and regime-dependent specifications for future work.

We could not run this ourselves. Our price data does not carry the JSE property indices and begins around 2010, six years after the sample starts. We also have no ASSA climate series. US REITs would test a different hypothesis because the climate channel depends on where the buildings are.

An EXMI-augmented model beating plain GARCH(1,1) out-of-sample on J253 and J254, with alpha + beta strictly below one, would change my view.