Forecasting in a Data-Rich Enviornment

With the introduction of new macroeconomic and financial indicators and the timely publication of high frequency data, forecasters face an ever-increasing amount of information when making their predictions. It is thus a great challenge to set up parsimonious time series models that can synthesize the rich information set at hand, as well as make accurate forecasts. I hope in my dissertation to contribute to the forecasting literature by applying newly-developed tools and methods to the empirical forecasting of macroeconomic and business indicators. Chapter 1 examines the information contained in financial market signals can be informative regarding the state of the macro economy. In this chapter, we utilize principal component analysis and forecast combination techniques to summarize the information from a large panel of 991 financial market series. We examine the consensus GDP and CPI projections in two surveys of professional macro forecasts for their efficiency regarding the aforementioned signals. Our results show that their forecast errors correlate significantly with many financial series as well as factors extracted from these series. Using a panel of financial market data, we were able to predict professional forecasters' errors out-of-sample, indicating the potential to improve their forecasts with a rich set of financial signals. In addition, both the in-sample correlation and the out-of-sample forecast improvement were shown to strengthen during the most recent financial crisis. In Chapter 2, we aim at designing statistical models to predict corporate earnings which either perform as well as, or even better than analysts. There are at least two challenges: (1) analysts use real-time data whereas statistical models often rely on stale data and (2) analysts use potentially large set of observations whereas models often are frugal with data series. In this chapter we introduce newly-developed mixed frequency regression methods that are able to synthesize rich real-time data and predict earnings out-of-sample. Our forecasts are shown to be systematically more accurate than analysts' consensus forecasts, reducing their forecast errors by 15% to 30% on average, depending on forecast horizon. In Chapter 3, we propose imposing structure on the coefficients of an autogressive (AR) model to reduce the number of parameters estimated, and show that with a finite sample, such hyper-parameterization can lead to a more parsimonious model and can thus improve the AR model's forecast performance. Monte Carlo simulations were carried out to assess under which conditions the models we propose outperform the benchmark AR model. In an empirical application of forecasting 170 monthly macroeconomic series, we found that hyper-parameterized AR models have clear advantage over the AR model, for series where the population best linear projections are long.Doctor of Philosoph