This thesis documents research and findings of three essays in the area of prediction and forecast of economic cycles. Each essay in this thesis is dedicated to address one particular aspect of the research. The thesis also contributes to the existing research by providing additional empirical evidence on the predictability of the real economic activity and recessions in the US. The first essay (Chapter 2) focuses on the long horizon inference methods, and examines the predictability of real GDP growth rate in the US using several well known predictors. A battery of specifically designed inference methods are employed in the analysis to address statistical complications introduced by overlapping long horizon dependent variable. The recursive moving block Jack-knife method is used to correct the biased estimated coefficients. The second essay (Chapter 3) puts emphasize on the out-of-sample forecast evaluation between nested and non-nested model. The forecast performances of various forecasting models are evaluated against two naive benchmark models, namely the random walk model and the autoregressive model. For the nested model, the asymptotically valid critical values for the forecast evaluation are derived from bootstrap simulations. The robustness of the test results are examined by Rossi and Inoue (2011 )'s robustness tests. The third essay (Chapter 4) utilizes the probit model to examine the predictability of recessions in the US. We evaluate the predictive power of several non-linear transformed predictors against that of the yield spread, and we also introduce a similar approach as in Rossi and !noue (2011) to examine the average and peak forecast ability of the predictors. The forecast performances of the predictors under various model specifications are carefully investigated in this chapter as well