The econometrics of discrete structural change and its applications in models of price overcharge
| dc.contributor.advisor | Boshoff, Willem H. | en_ZA |
| dc.contributor.author | Van Jaarsveld, Rossouw | en_ZA |
| dc.contributor.other | Stellenbosch University. Faculty of Economic and Management Sciences. Dept. of Economics. | en_ZA |
| dc.date.accessioned | 2021-03-05T10:34:55Z | |
| dc.date.accessioned | 2021-04-21T14:39:25Z | |
| dc.date.available | 2021-03-05T10:34:55Z | |
| dc.date.available | 2021-04-21T14:39:25Z | |
| dc.date.issued | 2021-03 | |
| dc.description | Thesis (PhD)--Stellenbosch University, 2021. | en_ZA |
| dc.description.abstract | ENGLISH SUMMARY : This dissertation studies unexplored time series issues related to tests of structural change and the modelling thereof. The practical application is centred around the estimation of price effects in collusive markets. The results, however, is more generally relevant, since structural change is inherent in many economic relationships. The dissertation is organized into three main chapters. The first part deals with the dating of structural breaks, i.e. determining the dates of structural breaks. The second part deals with the impact of non-stationarity when modelling structural changes. The third part deals with a practical application and provides guidance for practitioners. In cases where the date of the structural change is unknown, two econometric approaches can be followed. One method, is to apply a regime-switching or time varying-parameter model. The second approach is to use structural break tests to determine the break dates, and subsequently construct a dummy variable to control for the breaks in a regression model. In the first part, we investigate how various structural break tests can translate into parameter bias. The results are contrasted to regime-switching and time varying-parameter models to better understand the conditions under which these two approaches can be used interchangeably. Given that structural change may take various forms, we investigate the comparative performance of multiple techniques on different forms of structural change. Specically, we evaluate various forms of changes in the mean of the data generating process. To assess the performance of these methods, we rely on simulation evidence. Related, we also discuss complications of numerical optimization techniques and regime-switching models in simulation studies. The results are discussed in the framework of overcharge, where parameter estimates play a central role in the punishment of cartels. The second part deals with the impact of nonstationarity on dummy variable coecients. This part provides evidence for the conditions under which the distribution of the dummy variable parameter will differ significantly from the t-distribution. This is important since time series models are used in civil litigations by plaintiffs who are required to prove that the cartel had a significant impact on prices. Congruent with the first part, we illustrate how incorrect conclusions can be reached when the dummy variable is misspecified and t-statistic are used to draw inference. In other words, when the break dates are misdated and the nonstationary nature of the data caused the distribution of the dummy variable parameter to differ from the t-distribution. Related, we show how cointegration tests are influenced when the break dates are misspecified. We provide a discussion of the extent to which error correction modelling can be used to address some of these issues and emphasize specification problems that are specific to the estimation of overcharge. To demonstrate the practical insights of the simulation studies in the first and second part, I apply the various techniques to a European competition case. This part also provides guidance for practitioners. | en_ZA |
| dc.description.abstract | AFRIKAANSE OPSOMMING : Die proefskrif bestudeer probleme in tydreeks analise wat met toetse vir strukturele verandering, en die modellering daarvan, verband hou. Die praktiese toepassing fokus op die beraming van prysoorverhaling in markte waar mededingers saamspan. Buiten hierdie praktiese fokus, is die resultate van algemene belang, aangesien 'n verskeidenheid van ekonomiese verwantskappe dikwels deur struktuurveranderinge geraak word. Die proefskrif bestaan uit drie dele. Die eerste deel handel oor die datering van strukturele breke, met ander woorde die bepaling van die datums waarop die strukturele verandering neerslag in ekonomiese verwantskappe vind. Die tweede deel handel oor die effekte van nie stationer in modelle van strukturele verandering. Die derde deel bespreek 'n praktiese toepassing. In gevalle waar die datums van die breekpunte onbekend is, kan een van twee ekonometriese benaderings gevolg word. Een metode is om 'n regimewisselling model of tydwisselende parameters model te gebruik. 'n Tweede metode is om 'n strukturelebreek toets uit te voer ten einde die breekdatums vas te stel en dan 'n fopveranderlike, o.g.v. die resultate van hierdie datums, te skep. Hierdie fopveranderlike kan dan gebruik word om vir die strukturele breuke in 'n regressiemodel te kontroleer. In die eerste deel van die proefskrif evalueer ons hoe verskeie strukturele breekpunt toetse parameter sydigheid kan veroorsaak. Die resultate word met regimewisselling en tydwisselende parameter modelle vergelyk ten einde te bepaal onder watter omstandighede die opsomming van verskillende metodes uitruilbaar is. Gegewe dat strukturele verandering verskeie vorme kan aanneem, word alternatiewe tegnieke vir verskillende vorme van strukturele verandering ondersoek. Die soeklig val op verskeie gevalle waar die verandering in die gemiddelde van die datavoortbrengende proses is. Om die gedrag van verskeie metodes te evalueer word daar van simulasie gebruik gemaak. Die proefskrif bevat ook 'n bespreking van verwante probleme in verband met numeriese optimering en regimewisselling modelle in simulasie studies. Die resultate word bespreek in die konteks van prys oorverhalings waar parameter skattings 'n belangrike rol speel in die vergelding van kartels. Die tweede deel handel oor die impak van nie stationer op die koffesiente van fopveranderlikes in regressiemodelle. Hierdie deel omvat bewyse vir die omstandighede waaronder die verdeling van die fopveranderlike koffesient aansienlik van 'n t-verdeling verskil. Dit is belangrik aangesien tydreeksmodelle gebruik word deur klaers in siviele litigasie om bewyse te lewer dat samespanning 'n betekenisvolle impak op pryse gehad het. Ooreenkomstig met die eerste deel, wys hierdie deel hoe foutiewe gevolgtrekkings kan volg wanneer die fopveranderlike verkeerd gespesiseer word, dit wil se, wanneer die breekpunte verkeerd gedateer word en die nie stationere eienskappe van die tydreeks veroorsaak dat die fopveranderlike se parameterverdeling verskil van 'n t-verdeling. Die verwante probleem van hoe kointegrasie toetse beinvloed word wanneer die breekpuntdatums verkeerd bepaal is, word ook ondersoek. Die proefskrif bespreek voorts hoe foutkorreksie modelle gebruik kan word om van die hierdie probleme aan te spreek en beklemtoon spesikasieprobleme wat spesiek tot die beraming van prysoorverhaling is. Om die praktiese insig van die simulasie studies in die eerste en tweede deel te illustreer pas ek die verskeie metodes toe op 'n Europese kompetisie geval. In die deel verskaf ek ook 'n raamwerk vir praktisyns. | af_ZA |
| dc.description.version | Doctoral | |
| dc.format.extent | xiv, 173 pages ; illustrations, includes annexures | |
| dc.identifier.uri | http://hdl.handle.net/10019.1/110081 | |
| dc.language.iso | en_ZA | en_ZA |
| dc.publisher | Stellenbosch : Stellenbosch University | |
| dc.rights.holder | Stellenbosch University | |
| dc.subject | Econometrics | en_ZA |
| dc.subject | Structural change -- Statistical methods | en_ZA |
| dc.subject | Parameter estimation | en_ZA |
| dc.subject | Markup -- Statistical methods | en_ZA |
| dc.subject | UCTD | |
| dc.title | The econometrics of discrete structural change and its applications in models of price overcharge | en_ZA |
| dc.type | Thesis | en_ZA |