Macroeconomic forecasting: an examination of Australian markets
This thesis forms an empirical study on the forecasts of Australian macroeconomic variables. The forecast data comprises the period 1998-2018 and is taken from the Bloomberg survey of forecasters. The thesis has three main aims: firstly, to investigate the accuracy of forecasts of Australian macroeconomic data. Secondly, to examine whether forecasts are systemically biased from a behavioural perspective, specifically looking at forecaster herding and anchoring. And thirdly, to explore how forecast errors affect the Australian dollar and Australian 3- and 10-year bond futures markets. In the first section, we investigate forecast accuracy using a range of different statistical accuracy measures and evaluate whether the mean or the median is the superior consensus. We follow this with traditional forecast evaluation tests including the Pesaran-Timmermann (1992) test for directional accuracy, the Mincer-Zarnowitz (1969) test for unbiasedness, and the Elliot-Komunjer-Timmermann (2005) test for rationality. Consensus forecasts are then compared to a range of other forecast techniques, ranging from simple methods such as moving averages to more advanced methods such as neural networks and forecast combinations. We find strong evidence of correct directional accuracy and mixed evidence of forecast bias and rationality. Overall, we find minimal difference between mean and median forecasts. Additionally, consensus forecasts are shown to be a very difficult benchmark to beat. In the second section, we evaluate whether forecasts are behaviourally biased using the Bernhardt et al. (2006) test of herding. We initially examine herding behaviour before applying the test to anchoring. Numerous choices of anchor are implemented including some novel choices. We find evidence of anti-herding rather than herding which suggests that forecasters attempt to stand out with their forecast rather than blend in. We find mixed evidence for anchoring depending on the choice of anchor. We further examine the effect of market sentiment on herding and anchoring and find mixed results. In the third section, we explore the impact of these forecast errors on Australian financial markets, specifically looking at the Australian dollar and the 3- and 10-year bond futures markets within a 30 minute window surrounding the release. Data for this section is obtained from Thomson Reuters Tick History. The study is the most recent Australian study to examine forecast errors, referred to as news surprises in this context. We employ the Balduzzi et al. (2001) methodology to define news before estimating OLS regressions to examine how news affects markets. We investigate the asymmetric impact of positive versus negative news and the interaction between forecaster dispersion and news surprises. We find limited evidence of forecaster dispersion affecting market reaction but uncover time-variation in market response to news surprises.