Nigerian renewable-energy EGARCH leaves CPU out of volatility

A trader cannot read this CPU estimate as evidence that climate policy uncertainty drives volatility in Nigerian renewable-energy stocks. The paper estimates exactly one CPU coefficient, and it appears in the return equation. Its headline claim about asymmetric volatility has no CPU parameter behind it.

The Nigerian result

Owan and Asu have a plausible starting point: subsidies and regulation affect renewable-energy firms' earnings, so uncertainty about climate policy could hurt prices and raise volatility. The work they cite, Athari and Kirikkaleli, 2024 and Alharbey and Ben-Salha, 2024, reports that link in other markets. Their addition is a Nigerian sample fitted with an EGARCH(1,1) volatility model and Gaussian errors.

They calculate log returns for a renewable-energy series and use a climate policy uncertainty (CPU) index to explain them. The index comes from "databases developed by Gavriilidis and other international policy uncertainty repositories", according to the paper. In the mean equation, return = constant + b1 × CPU + error. The variance equation models log variance through its own lag, β, the size of the preceding standardized residual, α, and its sign, γ. The estimated CPU slope is -0.0028 (z = -3.11, p = 0.0019), alongside a constant of 0.0185. For variance, the estimates are β = 0.914, α = 0.287 and γ = -0.176 (z = -4.29). Owan and Asu interpret γ as showing that "negative climate policy uncertainty shocks" move volatility more than positive ones. They conclude that CPU "constitutes a significant risk factor" for the sector.

We did not find a sample period, data frequency or observation count in the paper. Its account of the dates is "recent years characterized by significant developments in climate policy". Nor did we find the stocks or index used for returns. The paper gives no construction details for the Gavriilidis index. To our knowledge, it measures US climate policy uncertainty through US news; the paper describes no Nigerian version.

Where does CPU enter?

The stated null hypothesis says CPU "has no significant effect on the volatility of renewable energy stock returns". The rejection relies on two p-values: p = 0.0019 for CPU in the mean equation and p = 0.0000 for γ. Neither tests that null. CPU changes the conditional mean in the reported model. The variance equation printed in the paper and its results table has ω, β, α and γ, with no CPU term.

The γ term multiplies ε(t-1)/√h(t-1), a standardized return innovation. At γ = -0.176, a fall in the renewable series raises next-period log variance more than an equally sized rise. The paper itself calls γ a leverage parameter. Its effect requires no CPU term. Calling this a response to "negative climate policy uncertainty shocks" creates another problem: a negative shock to an uncertainty index means less uncertainty, presumably good news.

A direct test would put CPU in the ln(h) equation and report its coefficient.

The implied return is -0.39

At the reported CPU mean of 145.23, the printed mean equation gives 0.0185 - 0.0028 × 145.23, or about -0.39 per period. The reported mean return is 0.012. Demeaning or rescaling CPU before estimation could reconcile those figures, but the paper says nothing about doing either. As printed, they conflict.

Size raises a separate question. A one-standard-deviation CPU change of 35.61 moves predicted return by about -0.10, more than the return standard deviation of 0.087. The reported correlation, -0.421, and the two standard deviations imply an OLS slope near -0.0010. The EGARCH slope is about 2.7 times as large. Variance weighting can change the estimate; the paper does not explain a factor of 2.7.

The data section names exchange rate, inflation and crude oil price as controls. The scope section says they "may be incorporated as control variables where appropriate". Yet Table 5 has only a constant and CPU in the mean equation. In Nigeria, oil and the naira are obvious candidates for what CPU might be capturing. CPU enters contemporaneously, without a lag.

A fit, without a named series

The diagnostics stand up on their own terms. Before estimation, ARCH-LM reports F = 12.754 (p = 0.0001). Afterwards, the residuals pass Ljung-Box Q(12) = 11.834 (p = 0.462), ARCH-LM = 1.228 (p = 0.271) and Jarque-Bera = 2.145 (p = 0.342). Raw-return kurtosis is 4.215. Together, those results suggest the Gaussian EGARCH absorbs this series' clustering and fat tails. With β = 0.914, volatility has an implied half-life of about 7.7 periods. The paper leaves open whether those periods are days or months.

The methodology still says checks "will be conducted". It promises a Phillips-Perron test and an AIC/SBC comparison, while reporting only ADF results (RER -7.342, CPU -5.864, against -2.891). We found one estimated specification.

We could not rerun the analysis. Our data does not cover Nigerian equities or a Nigerian climate policy uncertainty measure, and the paper never names the return series needed to rebuild it. Its contribution over the cited work is Nigerian evidence; without a named series and dates, that evidence cannot be checked.

A re-estimate using a named series and stated dates, with CPU in the variance equation, would change my view if that coefficient were significant. Until then, the CPU finding is a mean-equation correlation with an implied average return of -0.39. The reported asymmetry belongs to returns.