. 2021 Aug 16;7(1):61. doi: 10.1186/s40854-021-00277-7

Table 2.

The effect of oil prices on Japanese stock returns

	Panel A: OLS estimator			Panel B: WN-FGLS estimator
Sample periods	Model with no controls	Model 1	R² (%)	Model with no controls	Model 1	R² (%)	Model 2 (seasonal)	R² (%)
Full-sample	0.0161** (1.9865)	0.0168** (2.0599)	0.77	0.0201** (1.9911)	0.0206**(2.0404)	0.80	0.0210** (2.0884)	0.84
COVID-19 sample	0.0072*** (6.8010)	0.0086*** (6.6904)	1.96	0.0106*** (4.6034)	0.0126*** (5.5035)	2.14	0.0138*** (4.7074)	1.28
Pre-COVID-19 Sample 1	0.1114*** (6.7405)	0.1137*** (6.8842)	3.63	0.1664*** (7.9654)	0.1691*** (7.8464)	4.31	0.1696*** (7.8641)	4.36
Pre-COVID-19 Sample 2	0.0453** (2.0306)	0.0533** (2.2773)	0.47	0.1260*** (3.6939)	0.1339*** (3.6983)	2.74	0.1317*** (3.4613)	2.76

This tables reports predictability test results based on the following time-series regression (Model 1):

$R_{t} = α + β_{1} G O P_{t - 1} + β_{2} R_{t - 1} + β_{3} R V_{t - 1} + β_{4} M O N_{t} + β_{5} T U E_{t} + β_{6} T H U_{t} + β_{7} F R I_{t} + ε_{t}$

The second model which we refer to as Model 2 is of the form:

$R_{t} = α + β_{1} G O P_{t - 1} + β_{2} R_{t - 1} + β_{3} R V_{t - 1} + β_{4} M O N_{t} + β_{5} T U E_{t} + β_{6} T H U_{t} + β_{7} F R I_{t} + β_{8} J A N_{t} + β_{9} J A N_{t} + β_{10} F E B_{t} + β_{11} M A R_{t} + β_{12} A P R_{t} + β_{13} M A Y_{t} + β_{14} J U N_{t} + β_{15} A U G_{t} + β_{16} S E P_{t} + β_{17} O C T_{t} + β_{18} N O V_{t} + β_{19} D E C_{t} + ε_{t}$

where $R_{t}$ is the Japanese stock market returns (log percentage returns of the Nikkei price index); GOP is the growth rate in the WTI spot price of oil; $RV$ is the Nikkei stock average volatility index; MON, TUE, THU, and FRI are the day-of-the-week (Monday, Tuesday, Thursday and Friday) dummy variables and JAN, FEB, MAR, APR, MAY, JUN, AUG, SEP, OCT, NOV and DEC are to capture the seasonal effects respectively. The models are estimated using OLS with standard errors corrected using the Newey and West (1987) procedure such that the estimates are autocorrelation and heteroskedasticity consistent (Panel A). We also estimate the model using the Westerlund and Narayan (2015) flexible generalized least squares (WN-FGLS) estimator which makes the estimates heteroskedasticity, persistency and endogeneity consistent (Panel B). We only report the main slope coefficient relating to $β_{1}$ = 0 that examines the null hypothesis that GOP does not predict stock returns. Four sample periods are considered: the full sample period covers 01/04/2010 to 03/17/2021; the COVID-19 sample has data for the 12/31/2019 to 03/17/2021 period; the pre-COVID-19 Sample 1 covers the 01/04/2010 to 12/30/2019 period; and the pre-COVID-19 Sample 2 has data for the 10/01/2018 to 12/30/2019 period. Lastly, ** (***) denote statistical significance at the 5% (1%) level.