On the (nonlinear) relationship between exchange rate uncertainty and trade — An investigation of US trade figures in the group of seven

In this paper bilateral models formalizing monthly growth of US imports and exports are employed to investigate the potential of nonlinear relationships linking exchange rate uncertainty and trade growth. Parametric linear and nonlinear as well as semiparametric time series models are evaluated in terms of fitting and ex ante forecasting. The overall impact of exchange rate variations on trade growth is found to be weak. In periods of large exchange rate variations, trade growth forecasts gain from conditioning on volatility. Empirical results support the view that the relationship of interest might be non-linear and, moreover, lacks homogeneity across countries and imports vs. exports. JEL no. C14, C22, F31, F41     

Testing for non-linear dependence in inter-war exchange rates

Testing for Non-Linear Dependence in Inter-War Exchange Rates. — This paper tests weekly inter-war floating exchange rate data for the pound-dollar, pound-franc and pound-reichsmark for non-linearity. Initial tests reveal strong evidence of generic non-linearity in these series and indicate neglected non-linear structure in the residuals of linear representations. Attempts to model this structure using GARCH residual processes have only been partially successful. Thus, two parametric models of such non-linearity were estimated. Comparing the forecasts from these models shows the mean square forecast errors of linear-GARCH and bilinear models to be lower than those from linear forecasts for all series, and that SETAR model forecasts outperform all other models for the pound-dollar.     

Forecasts for international financial series with VMD algorithms

Recent models with variational mode decomposition (VMD) have been applied to time-series forecasting. In this paper, we build a hybrid model named VMD–autoregressive integrated moving average (ARIMA)–Taylor expansion forecasting (TEF) to increase accuracy and stability for predicting financial time series. We use VMD algorithms to decompose financial series into subseries. An ARIMA model is built to predict each mode’s linear component, and the pragmatic TEF model based on a tracking differentiator is applied to forecast of the nonlinear component. Then the forecasts of all subseries are assembled as a final forecast. Our empirical results of international stock indices demonstrate that the proposed hybrid approach surpasses several existing state-of-the-art hybrid models.     

FORECASTING PERFORMANCE OF AN ESTIMATED DSGE MODEL FOR THE SOUTH AFRICAN ECONOMY

We construct a small open‐economy New Keynesian dynamic stochastic general equilibrium (DSGE) model for South Africa with nominal rigidities, incomplete international risk sharing and partial exchange rate pass‐through. The parameters of the model are estimated using Bayesian methods, and its out‐of‐sample forecasting performance is compared with Bayesian vector autoregression (VAR), classical VAR and random‐walk models. Our results indicate that the DSGE model generates forecasts that are competitive with those from other models, and it contributes statistically significant information to combined forecast measures.     

The Incremental Informativeness of Stock Prices for Future Accounting Earnings*

This study extends previous research that documents a stock price reaction leading accounting earnings. The primary issue is that prior studies use a naive earnings expectation model (random walk) as the benchmark for the information content of lagged returns and do not adequately address the “incremental” information content of lagged returns. This study identifies and estimates firm-specific models of earnings to control directly for the autocorrelation in earnings. The explanatory power of lagged prices with respect to this earnings residual is investigated using both a multiple regression model of lagged returns and a multiple time-series vector autoregressive model. In-sample estimation of the models provides clear evidence that stock prices impound information about future earnings incremental to the information contained in historical earnings data. Holdout period analysis of the earnings forecasts from these lagged return models finds that both models outperform the naive seasonal random walk expectation, but neither model outperforms the more sophisticated Box-Jenkins forecasts. On an individual firm basis, earnings forecasts supplemented with the lagged return data tend to be less precise than the Box-Jenkins forecasts, but the price-based models demonstrate an ability to rank the earnings forecast errors from the time-series models. The analysis helps to characterize the limitations of lagged returns as a means of predicting future earnings innovations.     

Mixed-frequency approaches to nowcasting GDP: An application to Japan

In this paper, we discuss the approaches to nowcasting Japan’s GDP quarterly growth rates, comparing a variety of mixed frequency approaches including a bridge equation approach, Mixed-Data Sampling (MIDAS) and factor-augmented version of these approaches. In doing so, we examine the usefulness of a novel sparse principal component analysis (SPCA) approach in extracting factors from the dataset. We also discuss the usefulness of forecast combination, considering various ways to combine forecasts from models and surveys. Our findings are summarized as follows. First, some of the mixed frequency models discussed in this paper record out-of-sample performance superior to a naïve constant growth model. Second, albeit small, the SPCA approach of extracting factors improves predictive power compared with traditional principal component approach. Furthermore, we find that there is a gain from combining model forecasts and professional survey forecasts.     

An Analysis of Historical and Future-Oriented Information in Accounting-Based Security Valuation Models*

The Ohlson (1995) and Feltham and Ohlson (1995) valuation model provides a rigorous framework for summarizing the information in expected future earnings and book values. However, the model provides little guidance on selecting an empirical proxy for expected future earnings. We examine whether and under what circumstances historical earnings and analyst earnings forecasts offer comparable explanation of security prices. This issue is of particular interest because analyst forecasts are less readily available than historical data. Under appropriate circumstances, historical data may allow wider use of the Feltham-Ohlson valuation model by researchers and investors. A related issue is the incremental explanatory power of historical earnings and realized future earnings (perfect-foresight forecasts) for security prices beyond analyst forecasts. If historical earnings are incrementally informative, that would suggest that analyst forecasts do not fully reflect price-relevant information in past earnings. If future earnings are incrementally informative, that would suggest that security prices reflect investors' implicit earnings forecasts beyond analyst forecasts. We examine these issues using a historical model (based on past earnings), a perfect-foresight model (based on realized future earnings), and a forecast model (based on Value Line earnings forecasts). All three models provide significant explanatory power for security prices, and each set of earnings data provides incremental explanatory power for prices when used with the other sets of earnings data. We estimate the models separately for firms with moderate and extreme earnings-to-price (E/P) ratios, a proxy for earnings permanence. For moderate-E/P firms, the historical model's explanatory power exceeds that of the perfect foresight model, and is indistinguishable from that of the analyst forecast model. In contrast, for extreme-E/P firms, the perfect-foresight model offers greater explanatory power than the historical model, but lower explanatory power than analyst forecasts. Our results suggest that financial analysts' forecasting efforts are best focused on firms whose earnings contain large temporary components (extreme E/P firms). However, in general, both historical data and analyst forecasts are complementary information sources for security valuation.     

An Examination of the Statistical Significance and Economic Relevance of Profitability and Earnings Forecasts from Models and Analysts

In this paper, we propose and empirically test a cross‐sectional profitability forecasting model which incorporates two major improvements relative to extant models. First, in terms of model construction, we incorporate mean reversion through the use of a two‐stage partial adjustment model and inclusion of a number of additional relevant determinants of profitability. Second, in terms of model estimation, we employ least absolute deviation (LAD) analysis instead of ordinary least squares because the former approach is able to better accommodate outliers. Results reveal that forecasts from our model are more accurate than three extant models at every forecast horizon considered and more accurate than consensus analyst forecasts at forecast horizons of two through five years. Further analysis reveals that LAD estimation provides the greatest incremental accuracy improvement followed by the inclusion of income subcomponents as predictor variables, and implementation of the two‐stage partial adjustment model. In terms of economic relevance, we find that forecasts from our model are informative about future returns, incremental to forecasts from other models, analysts’ forecasts, and standard risk factors. Overall, our results are important because they document the increased accuracy and economic relevance of a cross‐sectional profitability forecasting model which incorporates improvements to extant models in terms of model construction and estimation.     

A BVAR MODEL FOR THE SOUTH AFRICAN ECONOMY

The paper develops a Bayesian vector autoregressive (BVAR) model of the South African economy for the period of 1970:1‐2000:4 and forecasts GDP, consumption, investment, short‐term and long term interest rates, and the CPI. We find that a tight prior produces relatively more accurate forecasts than a loose one. The out‐of‐sample‐forecast accuracy resulting from the BVAR model is compared with the same generated from the univariate and unrestricted VAR models. The BVAR model is found to produce the most accurate out of sample forecasts. The same is also capable of correctly predicting the direction of change in the chosen macroeconomic indicators.     

基于组合模型的股票预测研究综述

随着金融业数据环境的日益复杂,利用传统单一模型进行高精度股价预测变得愈加困难。面对日益突出的股票分析技术需求,组合预测模型开始得到发展并取得了很多成果。首先介绍影响股价波动的分析指标,概括基于传统统计预测模型、机器学习、神经网络等单一预测模型在股票预测中的优势与不足。然后依据组合预测模型的组合形式,将其分为线性模型的组合、非线性模型的组合以及线性与非线性模型的组合3种类型,并报告多种组合模型的实际应用与研究现状。最后,对组合模型股票预测方法的有效性和稳定性进行展望。     

Valuation Implications of Unconditional Accounting Conservatism: Evidence from Analysts' Target Prices

We examine whether financial analysts understand the valuation implications of unconditional accounting conservatism when forecasting target prices. While accounting conservatism affects reported earnings, conservatism per se does not have an effect on the present value of future cash flows. We examine whether analysts adjust for the effect of conservatism included in their earnings forecasts when using these forecasts to estimate target prices. We find that signed target price errors (actual minus forecast) have a significant positive association with the degree of conservatism in forward earnings, suggesting that target prices are biased due to accounting conservatism. Cross‐sectional analysis suggests that more sophisticated analysts and superior long‐term forecasters adjust for conservatism to a greater extent than other analysts. In additional analyses, we explore the mechanism through which conservatism leads to bias in target prices. We first show that analysts' earnings forecasts are negatively associated with the degree of conservatism; that is, analysts include the effect of unconditional conservatism in their earnings forecasts. Based on alternative earnings‐based valuation models that analysts may use, our evidence suggests that analysts fail to appropriately adjust their valuation multiple for the effect of conservatism included in their earnings forecasts when using these forecasts to derive target prices. As a consequence, we find that, for extreme changes in conservatism, the bias in analysts' target prices due to conservatism leads to a distortion of market prices. The evidence highlights the concern that analysts may not appreciate the valuation implications of conservative accounting which could inhibit price discovery.     

The Role of Current Account Balance in Forecasting the US Equity Premium: Evidence From a Quantile Predictive Regression Approach

The purpose of this paper is to investigate whether the current account balance can help in forecasting the quarterly S&P500-based equity premium out-of-sample. We consider an out-of-sample period of 1970:Q3 to 2014:Q4, with a corresponding in-sample period of 1947:Q2 to 1970:Q2. We employ a quantile predictive regression model. The quantile-based approach is more informative relative to any linear model, as it investigates the ability of the current account to forecast the entire conditional distribution of the equity premium, rather than being restricted to just the conditional-mean. In addition, we employ a recursive estimation of both the conditional-mean and quantile predictive regression models over the out-of-sample period which allows for time-varying parameters in the forecast evaluation part of the sample for both of these models. Our results indicate that unlike as suggested by the linear (mean-based) predictive regression model, the quantile regression model shows that the (changes in the) real current account balance contains significant out-of-sample information when the stock market is performing poorly (below the quantile value of 0.3), but not when the market is in normal to bullish modes (quantile value above 0.3). This result seems to be intuitive in the sense that, when the markets are performing average to well, that is performing around the median and above of the conditional distribution of the equity premium, the excess return is inherently a random-walk and hence, no information, from a predictor (changes in the real current account balance) is able to predict the equity premium.     

Evaluating CPB’s Forecasts

This paper analyzes forecasts, for ten key annually observed economic variables for the Netherlands, created by the Netherlands Bureau for Economic Policy Analysis (CPB) for 1971–2007. These CPB forecasts are all manually modified model forecasts, where the model is a (very) large multi-equation macro model. The CPB forecasts are held against real-time forecasts obtained from simple autoregressive time series models, and for seven of the ten cases, CPB’s forecasts are significantly more accurate. Combining CPB’s forecasts with the real time autoregressive forecasts shows that four of the ten combined forecasts are significantly better than CPB’s forecasts, and seven of the ten are better than the time series forecasts. This suggests that CPB’s manual adjustment efforts could perhaps be directed to modifying simple model forecasts and not the forecasts from the own large macro model.     

Conditional forecast selection from many forecasts: An application to the Yen/Dollar exchange rate

This paper proposes a new method for forecast selection from a pool of many forecasts. The method uses conditional information as proposed by Giacomini and White (2006). It also extends their pairwise switching method to a situation with many forecasts. I apply the method to the monthly yen/dollar exchange rate and show empirically that my method of switching forecasting models reduces forecast errors compared with a single model.     

Investors' Misweighting of Firm-Level Information and the Market's Expectations of Earnings

Prior studies use fundamental earnings forecasts to proxy for the market's expectations of earnings because analyst forecasts are biased and are available for only a subset of firms. We find that as a proxy for market expectations, fundamental forecasts contain systematic measurement errors analogous to those in analysts' biased forecasts. Therefore, these forecasts are not representative of investors' beliefs. The systematic measurement errors from using fundamental forecasts to proxy for market expectations occur because investors misweight the information in many firm-level variables when estimating future earnings, but fundamental forecasts are formed using the historically efficient weights on firm-level variables. Thus, we develop an alternative ex ante proxy for the market's expectations of future earnings (“the implied market forecast”) using the historical (and inefficient) weights, as reflected in stock returns, that the market places on firm-level variables. A trading strategy based on the implied market forecast error, which is measured as the difference between the implied market forecast and the fundamental forecast, generates excess returns of approximately 9 percent per year. These returns cannot be explained by investors' reliance on analysts' biased forecasts. Overall, our results reveal that market expectations differ from both fundamental forecasts and analysts' forecasts.     

Short Sellers and Long-Run Management Forecasts

We examine how short sellers affect long-run management forecasts using a natural experiment (Regulation SHO) that relaxes short-selling constraints on a group of randomly selected firms (referred to as pilot firms). We find that compared to other firms, the pilot firms issue more long-run good news forecasts but do not change the frequency of long-run bad news forecasts. The increase in good news forecasts is greater when the pilot firms have higher-quality forecasts, greater uncertainty about firm value, or higher manager equity incentives. Overall, these results and the results of additional analyses indicate that the reduction in short-selling constraints and the increase in short-selling threat induce managers to enhance disclosures through more long-run good news forecasts to discourage short sellers.     

An analysis of RSQE forecasts: 1971–1992

The purpose of this paper is to evaluate the accuracy of ex ante econometric model forecasts of four key macroeconomic variables: real GNP growth, the rate of price inflation measured by the GNP deflator, the civilian unemployment rate, and the Treasury Bill rate. Annual forecasts produced by the Research Seminar in Quantitative Economics (RSQE) based on the Michigan Quarterly Econometric Model of the U.S. Economy are compared with quasi ex ante forecasts from a four-variable vector autoregressive (VAR) model. Statistical tests of the equality of forecast error variances as well as univariate and multivariate forecast encompassing-type tests are conducted. The forecast error variance comparisons indicate that for three of the four variables the RSQE forecasts are more accurate than the VAR forecasts and for one of the variables (real GNP growth) only slightly less accurate. The forecast encompassing-type tests indicate that the RSQE forecasts contain information not contained in the VAR forecasts and, conversely, that VAR forecasts contain information not included in the RSQE forecasts. The scope for improving RSQE forecasts by combining them with VAR forecasts is rather limited, however.     

How Useful are DSGE Macroeconomic Models for Forecasting?

A review of the literature shows that forecasts from DSGE models are not more accurate than either times series models or official forecasts, but neither are they any worse. Further, all three types of forecast failed to predict the recession that started in 2007 and continued to forecast poorly even after the recession was known to have begun. The aim of this paper is to investigate why these results occur by examining the structure of the solution of DSGE models and compare this with pure time series models. The main factor seems to be the dynamic structure of DSGE models. Their backward-looking dynamics gives them a similar forecasting structure to time series models and their forward-looking dynamics, which consists of expected values of future exogenous variables, is difficult to forecast accurately. This suggests that DSGE models should not be tested through their forecasting ability.     

Extended Dividend,Cash Flow,and Residual Income Valuation Models: Accounting for Deviations from Ideal Conditions*

Previous empirical studies derive the standard equity valuation models (i.e., DDM, RIM, and DCF model) while assuming that ideal conditions, such as infinite payoffs and clean surplus accounting, exist. Because these conditions are rarely met, we extend the standard models by following the fundamental principle of financial statement articulation. We then empirically test the extended models by employing two sets of forecasts: (1) the analyst forecasts provided by Value Line, and (2) the forecasts generated by cross‐sectional regression models. The main result is that our extended models yield considerably smaller valuation errors. Moreover, by constructing these models, we obtain identical value estimates across the extended models. By reestablishing empirical equivalence under nonideal conditions, our approach provides a benchmark that enables us to quantify the errors caused by individual deviations from ideal conditions and thus to analyze the robustness of the standard models. Finally, by providing a level playing field for the different valuation models, our findings have implications for other empirical approaches, for example, estimating the implied cost of capital.     

A random coefficient model of speculative attacks: The case of the Mexican peso

A considerable body of theoretical and empirical literature has developed seeking to explain the timing, magnitude, and mechanics of speculative attacks against currencies. This paper extends the empirical specification of the traditional speculative attack model by developing a random coefficient (RC) model which, as we show, encompasses a variety of fixed-coefficient models as special cases. Two classes of models (fixed- and random-coefficient models) are estimated for the case of Mexican peso over the period January 1988 to Novemeber 1994, while forecasts of the peso/U.S. dollar exchange rate are generated for the period December 1994 through December 1995. The comparison of forecast errors generated by five model specifications indicates that forecasts based on the RC procedures are superior to those based on the fixed-coefficient estimation. It is also shown that there are good theoretical reasons why the RC procedure performs better in prediction than the fixed-coefficient procedure.