MIT Press

The first step is to examine the nature of the data. The presence of unit roots in most macroeconomic variables is fairly common (Nelson and Plosser 1982). Hence, estimating the long-run relationships of stationary variables using standard cointegration techniques (e.g., Johansen cointegration) is inconsistent. Therefore, unit root tests on the variables used in this paper are imperative. Conventional research has used both the Augmented Dickey–Fuller (ADF) (Dickey and Fuller 1979, 1981) and the Phillips–Perron (PP) (Phillips and Perron 1988) tests to identify the existence of unit roots. Elliott, Rothenberg, and Stock (1996) proposed the Dickey–Fuller Generalized Least Square (DFGLS) test, which is a modified version of the standard ADF test. According to the DFGLS procedure, the data are detrended before testing for stationarity. Different versions of the ADF, PP (with no constant and trend, constant and no trend, and constant and trend), and DFGLS tests (with constant but without trend, and constant and trend) are applied in this paper. All of these versions produce similar results. Due to space constraints, only the results with constant but without trend are presented here. All remaining results are available upon request.^² Unit root results for monthly variables are displayed in Table 2 and the results for quarterly variables are displayed in Table 3. For the monthly dataset, most variables are nonstationary at levels and stationary at the first difference. The year-on-year percentage change in consumer price index is found to be nonstationary at levels and stationary at the first difference by two out of three tests. The year-on-year percentage change in industrial production (IPIYOY) and the global bond market volatility index are stationary at levels. Thus, most variables are integrated of order one, I(1). All three tests suggest that IPIYOY is stationary at levels; that is, I(0). Similar results are found for the quarterly variables. Government bond as a percentage of GDP is found to be stationary at levels by the PP test, and nonstationary at levels by the ADF and DFGLS tests. Therefore, all quarterly variables are either I(0) or I(1).

Table 2.

Unit Root Tests for Monthly Variables

Variable	DFGLS	ADF	PP
IGB2YR	−1.29	−1.72	−1.86
ΔIGB2YR	−1.76^*	−11.57^***	−11.57^***
IGB3YR	−1.26	−1.81	−1.97
ΔIGB3YR	−2.01^**	−7.60^***	−11.54^***
IGB5YR	−1.26	−1.95	−2.03
ΔIGB5YR	−2.44^**	−7.87^***	−11.38^***
IGB7YR	−1.27	−2.06	−2.06
ΔIGB7YR	−2.74^***	−7.96^***	−11.18^***
TB3M	−1.57	−2.57	−2.58
ΔTB3M	−2.15^**	−17.09^***	−17.13^***
TCPIYOY	−1.63^*	−1.89	−1.99
ΔTCPIYOY	−9.47^***	−9.51^***	−9.48^***
IPIYOY	−1.92^*	−4.67^***	−13.66^***
ΔIPIYOY	−0.97	−9.73^***	−47.57^***
CREDIT	0.30	−1.54	−1.64
ΔCREDIT	−0.98	−2.48	−6.99^***
NEER	0.48	−0.52	−0.27
ΔNEER	−0.79^*	−11.21^***	−11.04^***
RISK	−4.93^***	−4.93^***	−4.86^***
ΔRISK	−0.97	−17.18^***	−19.01^***

Variable	DFGLS	ADF	PP
IGB2YR	−1.29	−1.72	−1.86
ΔIGB2YR	−1.76^*	−11.57^***	−11.57^***
IGB3YR	−1.26	−1.81	−1.97
ΔIGB3YR	−2.01^**	−7.60^***	−11.54^***
IGB5YR	−1.26	−1.95	−2.03
ΔIGB5YR	−2.44^**	−7.87^***	−11.38^***
IGB7YR	−1.27	−2.06	−2.06
ΔIGB7YR	−2.74^***	−7.96^***	−11.18^***
TB3M	−1.57	−2.57	−2.58
ΔTB3M	−2.15^**	−17.09^***	−17.13^***
TCPIYOY	−1.63^*	−1.89	−1.99
ΔTCPIYOY	−9.47^***	−9.51^***	−9.48^***
IPIYOY	−1.92^*	−4.67^***	−13.66^***
ΔIPIYOY	−0.97	−9.73^***	−47.57^***
CREDIT	0.30	−1.54	−1.64
ΔCREDIT	−0.98	−2.48	−6.99^***
NEER	0.48	−0.52	−0.27
ΔNEER	−0.79^*	−11.21^***	−11.04^***
RISK	−4.93^***	−4.93^***	−4.86^***
ΔRISK	−0.97	−17.18^***	−19.01^***

ADF = Augmented Dickey–Fuller, CREDIT = credit to the private sector as percentage of GDP, DFGLS = Dickey–Fuller Generalized Least Squares, IGB2YR = 2-year government bond yield, IGB3YR = 3-year government bond yield, IGB5YR = 5-year government bond yield, IGB7YR = 7-year government bond yield, IPIYOY = year-on-year percentage change in industrial production, NEER = nominal effective exchange rate, PP = Phillips–Perron, RISK = global bond market volatility index, TB3M = 3-month government auction rate, TCPIYOY = year-on-year percentage change in consumer price index.

Notes: ^***, ^**, and ^* indicate statistical significance at 1%, 5%, and 10% levels, respectively. The null hypothesis of all three tests is that the series contains unit roots.

Source: Authors’ calculations.

View Large

Table 3.

Unit Root Tests for Quarterly Variables

Variable	DFGLS	ADF	PP
IGB2YR_Q	−1.51	−2.05	−2.05
ΔIGB2YR_Q	−6.10^***	−7.47^***	−7.48^***
IGB3YR_Q	−1.60	−2.27	−2.14
ΔIGB3YR_Q	−6.36^***	−8.06^***	−8.36^***
IGB5YR_Q	−1.72^*	−2.54	−2.30
ΔIGB5YR_Q	−6.58^***	−8.51^***	−9.59^***
IGB7YR_Q	−1.81^*	−2.72	−2.47
ΔIGB7YR_Q	−6.77^***	−6.81^***	−10.14^***
TB3M_Q	−1.59	−2.16	−2.57
ΔTB3M_Q	−1.87^*	−8.52^***	−8.60^***
TCPIYOY_Q	−1.93^*	−2.36	−2.44
ΔTCPIYOY_Q	−6.46^***	−6.56^***	−6.65^***
IPIYOY_Q	−1.70^*	−4.64^***	−4.58^***
ΔIPIYOY_Q	−6.55^***	−6.53^***	−14.18^***
DRATIO_Q	−1.27	−2.21	−4.00^***
ΔDRATIO_Q	−0.87	−2.60^*	−11.21^***

Variable	DFGLS	ADF	PP
IGB2YR_Q	−1.51	−2.05	−2.05
ΔIGB2YR_Q	−6.10^***	−7.47^***	−7.48^***
IGB3YR_Q	−1.60	−2.27	−2.14
ΔIGB3YR_Q	−6.36^***	−8.06^***	−8.36^***
IGB5YR_Q	−1.72^*	−2.54	−2.30
ΔIGB5YR_Q	−6.58^***	−8.51^***	−9.59^***
IGB7YR_Q	−1.81^*	−2.72	−2.47
ΔIGB7YR_Q	−6.77^***	−6.81^***	−10.14^***
TB3M_Q	−1.59	−2.16	−2.57
ΔTB3M_Q	−1.87^*	−8.52^***	−8.60^***
TCPIYOY_Q	−1.93^*	−2.36	−2.44
ΔTCPIYOY_Q	−6.46^***	−6.56^***	−6.65^***
IPIYOY_Q	−1.70^*	−4.64^***	−4.58^***
ΔIPIYOY_Q	−6.55^***	−6.53^***	−14.18^***
DRATIO_Q	−1.27	−2.21	−4.00^***
ΔDRATIO_Q	−0.87	−2.60^*	−11.21^***

ADF = Augmented Dickey–Fuller, DFGLS = Dickey–Fuller Generalized Least Squares, DRATIO_Q = government debt as percentage of nominal gross domestic product, IGB2YR_Q = 2-year government bond yield, IGB3YR_Q = 3-year government bond yield, IGB5YR_Q = 5-year government bond yield, IGB7YR_Q = 7-year government bond yield, IPIYOY_Q = year-on-year percentage change in industrial production, PP = Phillips–Perron, TB3M_Q = 3-month government auction rate, TCPIYOY_Q = year-on-year percentage change in consumer price index.

Notes: ^***, ^**, and ^* indicate statistical significance at 1%, 5%, and 10% levels, respectively. The null hypothesis of all three tests is that the series contains unit roots.

Source: Authors’ calculations.

View Large

This Feature Is Available To Subscribers Only