Page 27 - Azerbaijan State University of Economics

THE Luan Vardari, Kiran Sood: Sate-Dependent Transmission of Oil and Electricity Shocks to
Equity Markets: Evidence from Emerging and Transitional Economies

To make the time series stationary, all the time series were converted into logarithmic
returns. Unit root tests ADF, PP and KPSS established that the level variables are I(1)
and the series of returns are I(0) which are stationary. Descriptive statistics indicated
that there were significant non-normality and excess kurtosis so nonlinear models and
regime-switching estimation procedures could be used.

3.2 Methodological Framework
The methodological approach builds on the nonlinear modeling structures developed
by Bildirici and Badur (2019) and Raifu and Oshota (2023). In this study, these
frameworks are integrated to form a unified empirical design that combines the Panel
Markov-Switching Vector Autoregressive (MS-VARX) model, Markov-Switching
Granger Causality, Dynamic Conditional Correlation GARCH (DCC-GARCH), and
copula-based dependence models. In this way, from this comprehensive framework,
a detailed examination of the links between these energy and financial factors is made,
which takes into account both structural modifications and nonlinear relationships that
are considered to vary on a market basis.

(a) Panel Markov Switching VARX
In order to account for potential regime transitions that are not immediately apparent,
the baseline specification attempts to capture the changing relationships between
financial and energy indicators. The MS-VARX framework makes it feasible to
analyze how the strength and direction of relationships among variables change over
time, as well as to identify discrete market phases, such as times of stability and
volatility.

, = + ∑ ℓ, , −ℓ + + , , ∼ (0, Σ )
,
,

ℓ=1

⊤
where , = [STOCK , FX , ELEC ] are endogenous variables,
,
,
,
, = [OIL , , RENEW , FIT , CPO ]are exogenous,
,
,
,

,
and ∈ {1,2} represents the low- and high-volatility regimes, which evolve

according to a first-order Markov process.
Lag length = 2was selected via AIC and HQC criteria.
Regime transition probabilities = ( = ∣ −1 = )were estimated using the

Expectation Maximization algorithm of Hamilton (1990) and Krolzig (1997).
Country specific intercepts control for structural heterogeneity.

27