A threshold cointegration analysis of Norwegian interest rates

摘要: In this thesis we generalize the Hansen and Seo test in R package tsDyn, which tests a linear cointegration model against two-regime threshold model, to case of three regimes alternative hypothesis. As Lagrange Multiplier (LM) statistic used tsDyn is different from LM described (2002), both these statistics, show that they are equal under certain conditions. The uses SupLM maximum when two thresholds vary over set all possible values. grid search algorithm, necessary maximizing statistic, also extended regimes, it rewritten such if value β given, really maximizes constraints specified by user. our empirical studies have examined thoroughly bivariate time series consisting monthly NIBOR rates maturities tomorrow next 12 months. When modeling series, find strong evidence for TVECM being superior VECM, out-of-sample forecasting SETAR gives much better prediction relation than AR model. testing three-regime cannot be rejected at any reasonable significance level. addition, how influential few outliers may removing them rerunning some statistical tests. Also, tested 66 pairs Norwegian interest cointegration, term spread each pair effects, i.e., as well We lot cointegrated pairs, approximately 50 % cases, cases univariate analysis. At last, simulate with data generation process, estimate simulated series. Thus, illustrate version detects optimal, close original simulation. expected, strongly rejected, there .