| spelling |
ndltd-OhioLink-oai-etd.ohiolink.edu-osu12612450432021-08-03T05:57:51Z Covariate unit root test under structural change and its application to the relation between income and consumption Shin, Sukha <p>This dissertation comprises four essays on unit root test under structural change. I propose a new unit root test under structural change which has gains in power over conventional tests and, using the new test, examine the implication of a structural change on the long run relationship between income and consumption.</p><p>While a data set over a long time period is preferred for more reliable testing for a unit root, it is more likely to contain a structural change due to an event such as war, oil price shock, or monetary policy regime change. If there is a structural change, it should be allowed for in a unit root test. Otherwise, the unit root test will be biased toward false acceptance of a unit root. However, if a structural change is allowed for, there is loss of power. Indeed, the Perron test, a conventional unit root test under structural change, has a pitfall of low power in finite samples.</p><p>The first essay of this dissertation proposes a new unit root test under structural change that utilizes the information in covariates, the variables related to the variable being tested. Asymptotic local power functions show that the power gains are sub¬stantial when the covariates are highly correlated with the variable being tested. In Monte Carlo experiments on finite samples, the new test (hereinafter, the covariate-Perron test) outperforms the Perron test in terms of power, without sacrificing the accuracy of size.</p><p>In the second essay, we discuss the lag length selection method for the covariate-Perron test which includes lagged terms of the first-differenced dependent variable and lagged terms of covariates in the estimation procedure. While the true lag lengths are unknown in practice, the finite sample performance of the covariate-Perron test as well as its asymptotic property can be largely affected by the choice of the lag lengths. We examine two asymptotically valid methods for the lag length selection and find that a sequential general-to-specific method has an advantage over an information criterion method in terms of size distortion.</p><p>The third essay extends the covariate-Perron test for an unknown break point. We consider two test statistics each of which is combined with a different method for the selection of a break point. Through Monte Carlo experiments, we find that there is a trade-off of size distortion and low power between the two methods. We recommend to use the method involving less severe size distortion. Though there is loss of power compared to the covariate-Perron test with a known break point, the power gains from using covariate is evident when the true break point is unknown. Unlike the covariate-Perron test with a known break point, the performance of the test with an unknown break point is affected by the magnitude of a break. However, this is not a serious problem with the break magnitude encountered in practice.</p><p>In the fourth essay, the implication of a structural change on the long-run rela¬tionship between income and consumption is investigated using the covariate-Perron test. Standard permanent income hypothesis implies that there is a cointegrating relationship between log income and log consumption with a cointegrating vector (1,-1)'. In other words, income and consumption grow in a balanced way keeping a ratio between them in the long run. Despite its intuitive appealing, this cointegrating relationship has been rejected except with U.S. data. The possibility of a structural change as an explanation for the rejection is explored. We derive a modified cointegrating relationship between income and consumption under structural change in the framework of a permanent income hypothesis and test the modified cointegrating relationship with post-war Japanese and U.S. data. Regardless of whether we assume an unknown break point or a known break point at 1973:4, the first oil price shock, we can find empirical evidence for the modified cointegrating relationship with Japanese data. By contrast, we fail to obtain empirical evidence for the modified cointegrating relationship with U.S. data whereas we can find the evidence for the conventional cointegrating relationship with no break.</p> 2002 English text The Ohio State University / OhioLINK http://rave.ohiolink.edu/etdc/view?acc_num=osu1261245043 http://rave.ohiolink.edu/etdc/view?acc_num=osu1261245043 unrestricted This thesis or dissertation is protected by copyright: all rights reserved. It may not be copied or redistributed beyond the terms of applicable copyright laws.
|
