Essays on interference for economic data with temporal and cross-sectional dependence

Chapter 1: A new exponential tilting test for weakly dependent data In this chapter we define a new test statistic for performing inference on a parameter from a weakly dependent time series process. This new statistic is asymptotically pivotal, and does not require an estimator of the long-run variance of the data generating process. The statistic is derived using the method of exponential tilting on subsample estimators of the parameter of interest. The subsamples make up a fully overlapped rolling window of the entire data sample, and an estimator of the parameter is computed for each subsample. The method for computing the subsample estimator is chosen by the practitioner; a general class of estimation methods that contains most of the methods commonly used in economic research is allowed. The limiting distribution of the test statistic is derived using the fixed-b asymptotic framework of Kiefer and Vogelsang (2005). The distribution is shown to only depend on the size of the window, and the critical values necessary for hypothesis testing were obtained by simulation and are provided in the appendix. The local asymptotic power of the new test is analyzed and is shown to be comparable to existing methods. Simulations show that the new test can have smaller size distortions than existing methods in some cases. The new test is applied to foreign exchange rate data and fails to reject the forward foreign exchange market efficiency hypothesis in contrast to most previous studies.;Chapter 2: A new exponential tilting test for linear panel models with fixed effects In this chapter we extend the method of subsample ET tests to linear panel data with fixed effects. Two different types of subsample estimators are considered. The first is based on a window of observations from an individual in the cross section, and the second is based on a rectangular window of the entire panel. We derive test statistics for testing hypotheses about regression parameters based on both types of subsample estimators. The nonstandard asymptotic distributions of the test statistics are derived using the fixed-b approach. These statistics are asymptotically pivotal under mild assumptions. The distributions are nonstandard, but critical values necessary for hypothesis testing are computed by simulation and reported in the appendix. Simulation studies compare the finite sample performance of the new tests to existing methods, and show that the new tests have better size properties. We also analyze a special case of the first type of statistic when the bandwidth is equal to the sample size. We derive a test somewhat analogous to cluster robust t-tests that is robust to arbitrary correlation within cluster if the clusters are independent. We provide critical values for this test that depend on the number of clusters using simulation methods, and compare the cluster ET test's performance to existing methods. We apply the new spatial correlation robust ET test to re-examine the question of whether or not no-fault unilateral divorce laws had an effect on the divorce rate. We find almost no evidence of an effect in contrast with some prominent recent studies.;Chapter 3: A test of the Fisher effect with fractional integration In this paper I model inflation and the nominal interest rate as a fractionally integrated, auto regressive, moving average (ARFIMA) process in order to test the theoretical proposition that nominal interest rates move one for one with inflation known as the Fisher effect. Using the necessary integration conditions, I test the Fisher effect hypothesis in eleven developed countries. For each country, I first estimate and test the order of integration of inflation and the nominal interest rate with a univariate ARFIMA model. In most of the countries, the estimated relative order of integration between the inflation and interest rate series directly rules out a Fisherian link between inflation and nominal interest rates. Because of the estimated orders of integration, testing for the Fisher effect in the three remaining countries (Belgium, Greece, and the US) requires the novel approach of applying the identification scheme of King and Watson (1997) to the fractionally differenced series. As in many of the previous studies, I find little support for the Fisher effect.