Eickmeier, Sandra (2007) On Economic Forecasting and International Linkages - Three Empirical Factor Studies. PhD thesis, Universität zu Köln.
This dissertation both examines the forecast performance of large-scale factor models (chapter II) and employs these models to investigate international economic comovements (chapters III and IV). Chapter II investigates the forecast performance of factor models using a meta-analysis. We summarize the seemingly ambiguous results from existing studies which evaluate the forecast performance of factor models, and we identify the determinants of forecast quality in factor models using a meta-analysis. The main focus of chapter II has been the relative forecast performance of large-scale dynamic factor models for real economic activity and inflation. This meta-analysis has the main advantage of being less prone to subjectivity regarding the choice of papers and results than narrative survey articles. Our analysis reaches several conclusions. First, forecasts can be improved if information from large datasets is exploited. This has been derived from our finding that factor models outperform smaller time series models. Alternative methods which are also able to exploit information from large datasets even outperform or are comparable to factor models. This is also supported by the result that the size of the dataset from which the factors are extracted positively affects the predictions. Second, the target variable itself alters the quality of factor forecasts. According to our analysis, factor models are relatively better at predicting US and euro-area than British variables, and they work relatively better on average for inflation than for output. Third, it can pay off to carefully specify the model. More complex factor estimation techniques by Forni et al. (2000) and Kapetanios and Marcellino (2004) are shown to be better at predicting output than the Stock and Watson (1998, 2002a) approach which is less demanding on the specification. Chapter III estimates the extent and the dynamics of the transmission of macroeconomic shocks from the US to Germany between 1975 and 2002. From a large dataset containing almost 300 German and US macroeconomic variables, we estimate common factors by means of the structural dynamic factor model suggested by Forni and Reichlin (1998). We then identify US supply and demand shocks, and we investigate the impact of these shocks on German macroeconomic variables and variables covering the various transmission channels. We finally decompose the forecast error of German GDP for the second half of the 1990s and the recession phase in the US in 2001. The analysis is one of the first applications of the structural factor model to an international macroeconomic topic. Also, the focus on transmission channels is new. A US demand shock leads to a significant rise of German GDP; the impact dies out after a year. German GDP also goes up after a US supply shock, but the impulse response is insignificant. The trade channel dominates. We are not able to draw clear conclusions on the role of financial markets and the confidence channel. A historical decomposition of German GDP finally shows that negative domestic factors overcompensated positive influences from the US between 1995 and 2000. In contrast, the US recession in 2001 was the main culprit for the German slump. Chapter IV is motivated by the observation that comovements at business-cycle and long-run frequencies among EMU-members are far from perfect, and there is still persistent heterogeneity. We establish stylized facts about comovements and heterogeneity of output and price developments in EMU member states. We combine the non-stationary dynamic factor approach of Bai and Ng (2004) and the structural factor setup based on Forni and Reichlin (1998) and apply them to a newly constructed partly non-stationary dataset containing a total of 173 quarterly macroeconomic times series from 1981 to 2003, which mainly capture economic developments in euro-area countries. Previous studies that examine economic linkages in the euro area fit stationary factor models to stationary datasets. The Bai and Ng (2004) approach allows us to examine comovements and heterogeneity without imposing restrictions on the persistence of the variables and their components (which are allowed to be non-stationary) and hence also on comovements and heterogeneity. Second, most of the previous studies are not concerned about the economic interpretation of the factors. Chapter IV investigates the transmission of a rich set of identified structural shocks. We find that individual euro-area countries� output and prices are hit by permanent common and idiosyncratic shocks. Idiosyncratic shocks and adjustments to them seem to be mainly responsible for cross-country heterogeneity throughout most of the sample period. The asymmetric transmission of common shocks seems to play a minor role. We find no strong evidence that some common shocks lead to greater cross-country heterogeneity than others.
Actions (login required)