That financial markets can influence real economic activity has been accepted by economists long ago and became dramatically apparent again in the last financial crisis, when the sharp decline in housing prices in the US was followed by a severe recession. In general, asset prices are determined in a forward-looking manner, stock prices for example reflect the expected profitability of firms in the future and thus are linked to expected future economic conditions. Furthermore, many macroeconomic models suggested by economic theory in-corporate interest rates, interest spreads or exchange rates, which can be seen as some sort of financial assets, and believing in these models means believing that asset prices influence developments of macroeconomic v ariables in the future. These considerations and observations gave rise to examine the pre-dictive power of asset prices to forecast output and inflation and a survey of this literature as well as empirical tests for a variety of predictors in different countries can be found e.g. in Stock and Watson (2003).
Table of Contents
1 Introduction
2 Methods
2.1 Pseudo Out-of-Sample forecasts
2.2 Benchmark forecasts
2.3 Forecast evaluation
2.4 Diebold Mariano test
2.5 Forecasting with individual predictors
2.6 Forecast combination
3 The Data
4 Results
4.1 Individual predictor forecasts
4.2 Combination forecasts
5 A (short) closer look at the last financial crisis
6 Conclusion
Objectives and Research Scope
This paper examines the predictive power of various financial assets for forecasting output and inflation, specifically addressing the instability of these predictors over time and across different countries.
- Analysis of individual predictor models using autoregressive distributed lag (ADL) frameworks.
- Evaluation of forecast combination techniques such as Equal Weighted Averaging (EWA) and Bayesian Model Averaging (BMA).
- Comparison of real-time forecasting performance using a pseudo out-of-sample exercise.
- Investigation of forecast failures during the 2008 financial crisis.
- Validation of empirical results using the Diebold-Mariano test for predictive accuracy.
Excerpt from the Book
2.1 Pseudo Out-of-Sample forecasts
Imagine a forecaster living in the second quarter of 1995 (1995:II) wants to forecast GDP in 1995:III. Therefore he/she uses data, e.g from 1978:IV, up to 1995:II to estimate a forecasting model and then produces the (out-of-sample) forecast for 1995:III. This methodology can be simulated and iterated quarter by quarter into the future, creating a sequence of pseudo out-of-sample forecasts. The estimation period may be held fixed in terms of the starting point or length, creating an expanding or rolling window respectively. With an expanding window more and more observations are included over time to estimate the model, which should lead to better results if the model is correctly specified and stable over time. When using a rolling window on the other hand the number of observations used to estimate the model is fixed, e.g. 120 beginning with data from 1978:IV to 1995:II to produce the forecast for 1995:III, then shifting the window one quarter into the future, using data from 1979:I to 1995:III to produce the forecast for 1995:IV, and so on. Choosing the optimal window size in the presence of structural breaks is not a trivial task and discussed for example in Pesaran and Timmermann (2007) or Clark and McCracken (2009).
While this procedure mimics actually done real-time forecasting, it misses a certain feature of many macroeconomic time series if a single dataset is used. First, figures for GDP for example are published with a lag of some month after the quarter to which they refer. Therefore a so called ragged edge occurs at the end of the most recent horizon. Second, data revisions are done over a period of several years. The first figures published by statistical agencies like the Bureau of Economic Analysis (BEA) are only preliminary and get revised over some month if new information is available. Additionally annual revisions may change the first estimates and every 5 years the BEA undertakes benchmark revisions where changing definitions of variables can have an impact.
Summary of Chapters
1 Introduction: Provides the motivation for using financial asset prices as predictors for macroeconomic variables and reviews existing literature.
2 Methods: Outlines the mathematical framework for out-of-sample forecasting, benchmark models, evaluation criteria, and combination methods.
3 The Data: Describes the selection of financial and macroeconomic variables and the transformation processes applied to achieve stationarity.
4 Results: Presents the empirical findings of the forecasting exercise for both individual predictors and model combinations regarding GDP and CPI inflation.
5 A (short) closer look at the last financial crisis: Discusses the failure of the examined forecasting models to predict the economic downturn during the 2008 financial crisis.
6 Conclusion: Summarizes that asset prices do not consistently predict macro variables and that simple model combinations often outperform sophisticated ones.
Keywords
Asset Prices, GDP Growth, CPI Inflation, Forecasting, Out-of-Sample, ADL Model, Diebold Mariano Test, Forecast Combination, EWA, BMA, Financial Crisis, Real-time Data, Model Averaging, Predictive Accuracy, Econometrics
Frequently Asked Questions
What is the core focus of this academic paper?
The paper investigates whether financial asset prices—such as stock prices, interest rates, and exchange rates—can provide reliable information for forecasting future output (GDP) and inflation (CPI).
What are the primary themes discussed in the work?
Central themes include the evaluation of individual forecasting models versus model combinations, the assessment of predictive accuracy in real-time, and the inherent instability of forecasting models across different time periods.
What is the specific goal of the research?
The primary goal is to perform an empirical exercise to test if asset prices offer superior forecasting performance compared to simple benchmark models like a random walk or an autoregressive process.
Which scientific methods are employed for the analysis?
The author uses pseudo out-of-sample forecasting, various model combination methods (Equal Weighted, Relative Efficiency, Bayesian Model Averaging, and Factor Augmented Autoregression), and the Diebold-Mariano test to assess forecast significance.
What topics are covered in the main body of the paper?
The main body covers the theoretical framework for generating and evaluating forecasts, the preparation of the dataset, the empirical results of individual versus combined models, and a specific case study on the 2008 financial crisis.
Which keywords best characterize this work?
Key terms include Asset Prices, Forecasting, Model Combination, GDP Growth, Inflation, Out-of-Sample, and Econometric Evaluation.
Did the study find any individual asset that reliably predicts GDP growth?
No, the study found that adding financial variables as predictors generally did not improve upon the performance of simple univariate benchmarks for GDP growth, and in some cases, it even worsened the results.
What lesson does the author draw from the last financial crisis?
The author notes that the tested models failed to indicate the economic downturn in 2008, consistently pointing toward a recovery even as the recession worsened, highlighting the difficulty of predicting structural breaks.
- Arbeit zitieren
- Gerret Halberstadt (Autor:in), 2014, On the informational content of asset prices for output (and inflation) forecasting, München, GRIN Verlag, https://www.grin.com/document/276378
