Analiza prognoz makroekonomicznych dla polskiej gospodarki z uwzględnieniem procesu rewizji danych

Publicly available macroeconomic forecasts have an impact on the expectations of all the agents in the financial markets and in the economy. The latter constitute the foundation of many theories and models, some of which were honoured with the Nobel Prize. Research around expectations is focused on the main two areas: firstly, understanding their nature and secondly, their impact on the decisions and behavior of the economic agents. Having forecasting data we can test models of the expectations formation, we can examine their properties and then based on the conclusions, try to understand their impact on the economic processes. This journey starts with the collection of the right data, through the inspection of the expectations formation models, the predictive accuracy of the forecasts and related uncertainty. The full understanding of the forecasting process would not be possible without the thorough analysis of the data: available when the predictions are prepared and used for their evaluation. These constitute the so-called real-time data. There is rich empirical literature related to the macroeconomic forecasting, but it mainly concerns developed economies. The spectrum shrinks, even more, when we consider research that takes into account revision processes. To my best knowledge, the topic has not yet been analysed for Poland and only a few studies have touched upon some single variables. In this study, I analyse two new datasets. Firstly, a real-time dataset for Poland constructed from the monthly Statistical Bulletins published by the Central Statistical Office of Poland, consisting of several hundred variables for the period since 1995. Secondly, a set of forecasts of the basic macroeconomic variables collected from nine publicly available sources. The analysis of two new datasets creates a unique cognitive value-added and provides answers to many questions related to the quality of the public statistics and their impact on our understanding of the economic processes and their modelling. Due to lack of similar studies for Poland, I define the research hypothesis as a list of the stylized facts collected from the research based mainly on the American data. They concern the revision processes, forecasts and their relations. The main purpose of this work is to analyse macroeconomic forecasts for the Polish economy and verify their basic properties. I extend the standard approach and concrete the conclusions, by relating the results to the real-time data and examining statistical properties of the applied tests, particularly in small samples and in the presence of the measurement errors. Research hypotheses are the reference points in the subsequent chapters and are directly referred to in the conclusions. It shall be stressed that I present many detailed results related to the individual categories of the analysed variables, which indicate potential recommendations for the economic modelling. To the best of my knowledge, such analyses have not been yet conducted for Poland and can provide valid recommendations for econometricians and economists who use public statistics on a daily basis. The dissertation consists of four chapters. In the first chapter, I embed the discussed problem in the economic theory and subject literature. I lead the reader from the fundamental foundations of the expectations theory to the mathematical models testing their assumptions with the empirical forecasting data. I outline the importance of the revision process for understanding the results and the entire forecasting process. This preliminary chapter ends with a list of stylized facts about the real-time data, forecasts and their relations. The second chapter is dedicated to the new real-time data set for Poland and contains its description and detailed analysis of the revision processes for several dozen macroeconomic variables. The study confirms that for many variables the revision process is non-trivial and it is often impossible to identify its systematic cause. Some revisions are proved to be systematic and predictable to a significantly greater extent than in a similar studies for the developed economies. It is an important observation, which shall affect our understanding of the predictability of the forecast errors. I also confirm that the scale of the revisions, their nature and the moments of publication depend on the data format. Based on the Polish data I am also able to positively verify many hypotheses regarding the impact of data revisions on the modelling process. Statistical procedures commonly used to evaluate properties of the data-generating process can give significantly different results depending on the data vintage, i.e. series after consecutive revisions. For a large set of the analysed variables, the total impact of the revision on the forecasting process exceeds 10%. It is not uncommon to observe variables, for which the impact exceeds 100%, or situations in which the vintage has a direct impact on the forecast sign. Understanding of these results by the forecasters could significantly improve the quality of their predictions. The third chapter concerns the set of actual economic expectations coming from the publicly available sources. I analyse the wide set of predictions for Poland for the process of expectations formation, their accuracy, embedded information sets, uncertainty and interrelatedness. I relate all these considerations to the data revision processes and both theoretically and empirically explain their impact on the results, with the aim to increase understanding of the impact that revisions have on the forecasting process. I confirmed the superiority of the survey forecasts over the simple econometric models, which proves the significant value-added of the expert judgement. There is an exception, which is systematically biased forecasts. I also confirmed the hypothesis that the conclusions regarding the quality of forecasts are strongly dependent on the revision process: the higher the level of the revisions, the higher the predictability of the forecast errors and their magnitude. I prove that often error signs vary across data vintages. Interestingly, this pattern is weaker for longer horizon forecasts, which may indicate that they are averaged across the historical revisions. In general, domestic institutions produce more accurate predictions for the Polish economy and therefore combining forecasts could yield improvements only for some, the least predictable variables, e.g. from the labor market. Analysis of the processes of expectations formation confirms lack of rationality for most of the variables, with few exceptions to the short-term forecasts. What is more, the extrapolative models, although simple and intuitively interpretable, generally do not explain well the processes behind Polish forecasts. I also confirmed that some of the business cycle related variables explain a significant portion of the forecast errors variability. Finally, the analysis of uncertainty related to the forecasting process reveals that disagreement among forecasters increases in longer horizons, but only for the unrevised variables. I also confirmed that for the forecasts with up to two years horizon, the interpersonal dispersion can well approximate the intrapersonal uncertainty and that it has some cyclical properties. It also turns out that the effect of revisions is much stronger in explaining the disagreement between forecasts than other variables known from the literature related to the business cycle, monetary policy, volatility on the financial markets or political uncertainty. The fourth chapter presents the results of the Monte Carlo simulations assessing statistical properties of the rationality and extrapolative tests. I run them in various scenarios including small samples and data with the measurement errors. The dissertation ends with the summary referring to the outlined research hypotheses. In most of the cases, I confirm that the conclusions for the Polish economy are similar to the stylized facts for the developed markets. I highlight a few divergent results for forecasts with different horizons. Utilization of the real-time data allowed me to draw many interesting implications complement to the initial research hypothesis. The wide spectrum of analysed variables implies that the results can be widely applicable in the empirical studies based on the Polish macroeconomic data and their predictions.