CI020 Room Beveridge Hall SPECIAL SESSION ON ADVANCES IN DYNAMIC FACTOR ANALYSIS Chair: Christopher Otrok CI0441: Time-varying spillovers
Presenter: Christopher Otrok, University of Missouri and FRB St Louis, United States
We develop a Bayesian Dynamic Factor model to measure cross-country spillovers in macroeconomic aggregates. The model is able to distinguish common shocks across countries from spillovers of country specific shocks. The key innovation is to allow for time variation in the spillover channels. Spillovers may be different in expansions and contractions. They also may vary with specific events, such as financial crises. We apply the model to European countries to measure the evolution of spillovers over time.
CI0446: Common and country specific economic uncertainty
Presenter: Haroon Mumtaz, Queen Mary University of London, United Kingdom
A factor model with stochastic volatility is used to decompose the time-varying variance of Macroeconomic and Financial variables into contribu-tions from country-specific uncertainty and uncertainty common to all countries. We find that the common component plays an important role in driving the volatility of nominal and financial variables. The co-movement in volatility of real and financial variables has increased over time with the common component becoming more important over the last decade.
CI1824: Forecasting fed funds target changes
Presenter: Michael Owyang, Federal Reserve Bank of St Louis, United States
Most interest rate rules are continuous functions of deviations of output from its potential and expected inflation from its target. In practice, central banks move the target rate in discrete increments and base their decisions on a wide-range of data. We estimate a dynamic ordered probit model of movements in the federal funds rate. In our model, monetary policy reacts to a large dataset that is summarized by small set of dynamic factors.
We then use the model for out-of-sample forecasting and evaluate these forecasts using methods unique to problems of classification.
CO536 Room Bedford FINANCIAL CONDITIONS INDICES Chair: Garry Young
CO0267: A financial conditions index using targeted data reduction Presenter: Garry Young, Bank of England, United Kingdom Co-authors:Simon Price, George Kapetanios
Financial conditions indices (FCIs) aim to summarise the state of financial markets. We construct two types of measure: a principal component of a medium sized set of relevant financial indicators and an alternative that takes information from a large set of macroeconomic variables weighted by the joint covariance with subsets of financial indicators, using multiple partial least squares (MPLS). Our approach aims to weight latent factors from a macroeconomic data set using information from financial variables. Both are useful for forecasting monthly GDP, but the MPLS based FCIs are superior to that based on the PC.
CO0544: Financial stress regimes and the macroeconomy Presenter: Ana Galvao, University of Warwick, United Kingdom Co-authors:Michael Owyang
The aim is to identify financial stress regimes using a model that explicitly links financial variables with the macroeconomy. The financial stress regimes are identified using a large unbalanced panel of financial variables with an embedded method for variable selection and, empirically, are strongly correlated with NBER recessions. The empirical results on the selection of financial variables support the use of credit spreads to identify asymmetries in the responses of economic activity and prices to financial shocks. We use a novel factor-augmented vector autoregressive model with smooth regime changes (FASTVAR). The unobserved financial factor is jointly estimated with the parameters of a logistic function that describes the probabilities of the financial stress regime over time.
CO0932: A high frequency measure of U.S. GDP with application to financial conditions indexes Presenter: Scott Brave, Federal Reserve Bank of Chicago, United States
Co-authors:Andrew Butters
We use the data underlying the Chicago Fed National Activity and National Financial Conditions indexes to construct a weekly measure of U.S.
GDP growth. A mixed-frequency dynamic factor model is posed which links cyclical movements in GDP growth to a small set of latent weekly real and financial activity factors. Collapsed dynamic factor methods are used to estimate the latent factors in addition to a latent time-varying mean (or trend) for GDP growth. A real-time out-of-sample forecasting exercise is then conducted to evaluate the ability of the model to forecast near-term GDP growth relative to prominent surveys of professional forecasters.
CO1220: Financial conditions in the Euro area: A narrative of the crisis and its consequences for the real economy Presenter: Hiona Balfoussia, Bank of Greece, Greece
Co-authors:Heather Gibson
We construct and present financial conditions indices (FCIs) for the Euro area, for the period 2003 onwards, using a wide range of prices, quantities, spreads and survey data, grounded in the theoretical literature. The FCIs fit in well with a narrative of financial conditions since the creation of the monetary union. FCIs for individual Euro area countries are also provided, with a view to comparing financial conditions in core and periphery countries. There is evidence of significant divergence both before and during the crisis, which becomes less pronounced when monetary policy variables are included in the FCI. However, the impact of monetary policy on financial conditions appears not to be entirely symmetric across the Euro area. We subsequently explore the relationship between financial conditions and real economic activity in the Euro area as a whole and for Greece in particular, by estimating the potential impact of the TLTROs on aspects of economic activity. Our results suggest that financial conditions do have a significant effect on economic activity and thus the TLTROs, to the extent that they are designed to improve financial conditions, will provide a boost to the real economy. A further extension explores the link between financial conditions and firm investment, and finds it significant.
CO480 Room SH349 ADVANCES IN FINANCIAL FORECASTING Chair: Ekaterini Panopoulou CO0357: Forecasting market returns: Bagging or combining
Presenter: Andrew Vivian, Loughborough University, United Kingdom Co-authors:Steven Jordan, Mark Wohar
The aim is to provide evidence on applying the bagging method to forecast stock returns out-of-sample for the G7 and a broad set of Asian countries for which there is little prior evidence. We focus on using the recently developed bagging method that explicitly addresses model uncertainty and parameter uncertainty. We are amongst the first to apply the bagging method to market return predictability and amongst the first to examine if bagging can generate economic gains. We find that, when portfolio weight restrictions are applied, bagging generally improves forecast accuracy and generates economic gains relative to the benchmark; bagging also performs well compared to forecast combinations in this setting. We also provide new evidence that the results for bagging cannot be fully explained by data mining concerns. Finally, we report that bagging generates economic gains in G-7 countries and overall these gains are highest for countries with high trade openness and high FDI. The potentially substantial economic gains could well be operational given the existence of index funds for most of these countries.
CO0986: A Bayesian non-parametric multiple quantile model for forecasting the asset return distribution Presenter: Evangelia Mitrodima, London School of Economics, United Kingdom
Co-authors:Jim Griffin, Jaideep Oberoi
We perform a Bayesian analysis using Markov Chain Monte Carlo (MCMC) methods and in particular, an Adaptive Metropolis Hastings algorithm to jointly model selected quantiles of the asset return distribution. Bayesian methodology is widely used in the literature concerning quantile regression, for improved estimation based on the Regression Quantile (RQ) criterion, by employing the Asymmetric Laplace likelihood. The Asymmetric Laplace distribution is a skew distribution, which offers a possible mathematical link between the minimization of the RQ criterion and the maximum likelihood theory. However, this does not address the underlying time-varying interdependence between individual quantiles in a natural way. An alternative is to use a non-parametric setting, as this seems more desirable under the Bayesian framework. To this end, we consider a histogram generated by the joint quantile model to approximate the density of asset returns with posterior inference for the parameters.
CO1118: Quantile forecast combinations in realised volatility prediction Presenter: Ekaterini Panopoulou, University of Kent, United Kingdom Co-authors:Ioannis Vrontos, Spyridon Vrontos, Loukia Meligkotsidou
Whether it is possible to improve realised volatility forecasts by conditioning on macroeconomic and financial variables is tested. We employ complete subset combinations of both linear and quantile forecasts in order to construct robust and accurate stock market volatility predictions. Our findings suggest that the complete subset approach delivers statistically significant out-of-sample forecasts relative to the autoregressive benchmark and traditional combination schemes. A recursive algorithm that selects, in real time, the best complete subset for each predictive regression quantile succeeds in identifying the best subset in a time- and quantile-varying manner.
CO426 Room Holden COMMON FEATURES IN MACROECONOMICS AND FINANCE Chair: Joao Victor Issler CC0365: Quantile factor models
Presenter: Jesus Gonzalo, University Carlos III de Madrid, Spain Co-authors:Juan jose Dolado, Liang Chen
A novel concept is proposed: quantile factor models, where a few unobserved common factors affect all parts of the distributions of many observed variables. A simple two step procedure is proposed to estimate the common factors and the quantile factor loadings. Uniform consistency and weak convergence results for the entire quantile factor loading processes are obtained. Based on these results, we show how to make inference of general forms in a location-scale shift factor model. Simulation results confirm the good performance of our estimators in small to moderate sample sizes.
CO1436: Local unit root and inflationary inertia in Brazil Presenter: Osmani Guillen, Ibmec and BCB, Brazil Co-authors:Wagner Gaglianone
The purpose is to study the persistence of Brazilian inflation using quantile regression techniques. To characterize the inflation dynamics we employ a Quantile Autoregression model. In this model, the autoregressive coefficient may assume different values, allowing testing the asymmetry hypothesis for the inflation dynamics. Furthermore, the model allows investigating the existence of a local unit root behavior. In other words, the model enables to identify locally unsustainable dynamics, but still compatible with global stationarity. In addition, the model can be reformulated in a more conventional coefficient notation, in order to reveal the periods of local nonstationarity. Another advantage of this technique is the estimation method, which does not require knowledge of the innovation process distribution, making the approach robust against poorly specified models. An empirical exercise with Brazilian inflation data and its components illustrates the methodology. As expected, the behavior of inflation dynamics is not uniform across different conditional quantiles. In particular, the results can be summarized as follows: (i) the dynamics is stationary for most quantiles of the sample period; (ii) the process is nonstationary in the upper tail of the conditional distribution; and (iii) the periods associated with local unsustainable dynamics can be related to those of increased risk aversion and higher inflation expectations.
CO1381: Risk assessment of the Brazilian FX rate
Presenter: Wagner Gaglianone, Central Bank of Brazil, Brazil Co-authors:Jaqueline Marins
We construct several multi-step-ahead density forecasts for the foreign exchange (FX) rate based on statistical, financial data and economic-driven approaches. The objective is to go beyond the standard conditional mean investigation of the FX rate and (for instance) allow for asymmetric responses of covariates (e.g. financial data or economic fundamentals) in respect to exchange rate movements. We also provide a toolkit to evaluate out-of-sample density forecasts and select models for risk analysis purposes. An empirical exercise for the Brazilian FX rate is provided. Overall, the results suggest that no single model properly accounts for the entire density in all considered forecast horizons. Nonetheless, the GARCH model as well as the option-implied approach seem to be more suitable for short-run purposes (until three months), whereas the survey-based and some economic-driven models appear to be more adequate for longer horizons (such as one year).
CO1437: Consumption-wealth ratio and expected stock returns: Evidence from panel data on G7 countries Presenter: Joao Victor Issler, Getulio Vargas Foundation, Brazil
Co-authors:Andressa Monteiro de Castro
Using a recent theoretical framework, we perform an empirical investigation on how widespread is the predictability of a modified consumption-wealth ratio – CAY – once we consider the set of G7 countries. The G7 countries represent more than 64% of net global consumption-wealth and 46% of global GDP at market exchange rates. We evaluate the forecasting performance of CAY using a panel-data approach, since applying cointegration and other time-series techniques is now standard practice in the panel-data literature. Hence, we generalize Lettau and Ludvigson’s tests for a panel of important countries. We employ macroeconomic and financial quarterly data for the group of G7 countries, forming an unbalanced panel. For most
countries, data is available from the early 1990s until 2014Q1, but for the U.S. economy it is available from 1981Q1 through 2014Q1. Results of an exhaustive empirical investigation are overwhelmingly in favor of the predictive power of CAY in forecasting future stock returns and excess returns.
CO384 Room Senate MULTIVARIATE METHODS FOR ECONOMIC AND FINANCIAL TIME SERIES Chair: Gianluca Cubadda CO0465: Real time mixed frequency VARs: Nowcasting, backcasting and Granger causality
Presenter: Alain Hecq, Maastricht University, Netherlands Co-authors:Thomas Goetz, Lenard Lieb
A previous mixed-frequency VAR is extended to the inclusion of a noncausal part. This allows us to consider the future of the variables and to focus more closely on data releases. We propose a method based on the LAD estimator for estimating causal-noncausal VARs for mixed-frequency series. Test statistics are computed using a bootstrap approach. For the European Union, a system with the quarterly growth rate of the real gross domestic product and three monthly growth rates of the industrial production index, reveals that a pure noncausal VAR is preferred by the data to the usual reduced form causal VAR. This leads to what we call backcasting causality instead of the well know concept of Granger causality.
Consequently, our modeling implies a new way to forecast time series in real time.
CO0582: Industrial development in the Italian regions, 1861-1913: New evidence Presenter: Stefano Fachin, Rome Sapienza, Italy
Co-authors:Francesca Di Iorio, Carlo Ciccarelli
The aim is to tackle the problem of applying the approximate factor model to spatial data using as a case study manufacturing industrial value added in the Italian regions from 1861 to 1913 from a recently released dataset at industry level. The application of the factor model to spatial data raises two questions usually not addressed: (i) the analysis of the loadings, as their distribution for the different factors over the spatial units may reveal important features of the data; (ii) modelling the errors, as the possibile presence of spatial dependence may be also very important. With respect to the first issue we develop a boostrap test for the hypothesis that the loadings are equal to a known matrix, reporting encouraging results of some simulation experiments. The second point is explored estimating spatial autoregressive panel models. Given the analogy with FAVARs, we propose to label the two-step strategy entailed by estimation of spatial error models on the de-factored residuals as FASEM (Factor Augmented Spatial Error Model).
CO0891: Adjustments of the effects of measurement errors using instrumental variables and mixed-models in cointegration analysis Presenter: Hanwoom Hong, Seoul National University, Korea, South
Co-authors:Sung Ahn, Sinsup Cho
The effects of measurement errors on the reduced-rank estimator and the cointegrating test of error correction models had been studied. It was shown that the asymptotic bias is present because of endogeneity caused by the measurement errors. Two methods are suggested to deal with the endogeneity. One method employs instrument variables and the other introduces a moving average term in the error correction model. It is investigated that asymptotic properties of the reduced rank estimators based on these two methods and is found that these estimators are no longer asymptotically biased. The asymptotic distribution of the likelihood ratio test for the cointegrating ranks based on these methods is also obtained.
Finally, small sample properties of the estimators and the test through a Monte Carlo simulation study is investigated.
CO1686: Volatility spillovers with multivariate stochastic volatility models Presenter: Yuliya Shapovalova, Maastricht University, Netherlands Co-authors:Michael Eichler
Co-movements in financial time series suggest presence of volatility spillover effects among financial markets. Understanding fundamentals behind this phenomena is important for portfolio managers and policy makers. Currently in the literature GARCH-type models is the dominating approach for detecting volatility spillovers. Inference is often based on notions of causality in mean and variance. We aim to analyse volatility spillovers using a more natural approach for volatility modeling: multivariate stochastic volatility models (MSVM). The structure of MSVM allows to test for causality in volatility processes directly, and in contrast to GARCH models causality in variance and causality in volatility do not coincide in this framework. However, due to presence of latent volatility processes estimation of this class of models is a difficult task, it has been shown that standard methods of estimation such as quasi-maximum likelihood and GMM do not perform sufficiently well. We use alternative methods of estimation: particle filters and variational Bayes. The notion of Granger-causality is used to test for causal links in volatility processes and Dirac spike and slab priors for model selection which gives us a fully Bayesian approach. Finally, based on our estimation and model selection we build graphical models for further graphical inference of causal structure.
CO380 Room Court BAYESIAN NONLINEAR ECONOMETRICS Chair: Roberto Casarin
CO0519: Nonparametric conditional Beta
Presenter: John Maheu, McMaster University, Canada Co-authors:Azam Shamsi
The effect of the market return on the value of systematic risk using a semiparametric multivariate GARCH model is investigated. We nonparamet-rically estimate the dynamic conditional beta without any restrictive assumption on the joint density of the data. This model captures movements in systematic risk over time, and we find that the time-varying beta of a stock nonlinearly depends on the contemporaneous value of excess market returns. The model is extended to allow nonlinear dependence in Fama-French factors. In general, in highly volatile markets, beta is almost constant, while in stable markets, the beta coefficient can be highly and asymmetrically dependent on the value of the market excess return.
CO0675: Bayesian panel Markov-switching model with mixed data sampling Presenter: Massimiliano Marcellino, Bocconi University, Italy
Co-authors:Roberto Casarin, Claudia Foroni, Francesco Ravazzolo
A Bayesian panel Markov-switching model with mixed data sampling (MIDAS) is proposed. We follow the unrestricted MIDAS approach, provide a Markov-chain Monte Carlo (MCMC) procedure for posterior approximation.
CO0842: Bayesian nonparametric sparse seemingly unrelated regression model Presenter: Luca Rossini, Ca Foscari University of Venice, Italy
Co-authors:Roberto Casarin, Monica Billio
Seemingly unrelated regression (SUR) models are used in studying the interactions among economic variables of interest. In a high dimensional setting and when applied to large panel of time series, these models have a large number of parameters to be estimated and suffer of inferential problems. We propose a Bayesian nonparametric hierarchical model for multivariate time series in order to avoid the overparametrization and overfitting issues and to allow for shrinkage toward multiple prior means with unknown location, scale and shape parameters. We propose a two-stage hierarchical prior distribution. The first two-stage of the hierarchy consists in a lasso conditionally independent prior distribution of the
Normal-Gamma family for the SUR coefficients. The second stage is given by a random mixture distribution for the Normal-Normal-Gamma hyperparameters, which allows for parameter parsimony through two components. The first one is a random Dirac point-mass distribution, which induces sparsity in the SUR coefficients; the second is a Dirichlet process prior, which allows for clustering of the SUR coefficients. We provide a Gibbs sampler for posterior approximations based on introduction of auxiliary variables. Some simulated examples show the efficiency of the proposed methods. We study the effectiveness of our model and inference approach with an application to macroeconomics.
CO1211: Bayesian estimation of multimodal density features applied to DNA and economic data Presenter: Herman van Dijk, Erasmus University Rotterdam, Netherlands
Co-authors:Nalan Basturk
Important theoretical and practical issues that involve distribution functions that have multimodal densities occur in several scientific fields: bioin-formatics, finance, and international economic growth. A Bayesian approach is proposed to estimate shape and other features of a multimodal density and the uncertainty around these values. The method is applicable for continuous and discrete data distributions. For continuous mul-timodal data, we show that estimates based on mixtures of normal densities with an unknown number of components provide a straightforward method to evaluate density features. For discrete such data, we propose a mixture of shifted Poisson densities with an unknown number of com-ponents. Mixture density estimates are obtained using simulation-based Bayesian inference with density features treated as random variables.
Highest posterior intervals around features are automatically obtained without any extra computational effort. For discrete data a novel version of a Reversible Jump Markov Chain Monte Carlo (RJMCMC) method is developed which is an adapted version of Green’s method. Instead of applying the more restrictive approach of choosing a particular number of mixture components using information criteria as in an earlier work, our method allows for an unknown number of components.
CO396 Room Gordon WAVELET METHODS IN ECONOMICS Chair: Marco Gallegati
CO0569: Firm ownership and provincial CO2-emissions in China Presenter: Fredrik NG Andersson, Lund University, Sweden Co-authors:Sonja Opper
Within only three decades China has emerged from one of the worlds poorest agricultural economies to a major manufacturing economy taking up the largest share of global carbon dioxide emissions. The Chinese economy capitalist transformation has been gradual and the present economy is a hybrid relying on private production a sizable state-owned sector. A large literature has shown that private firms are economically more efficient than state-owned firms, but whether private firms also are more carbon efficiency is another question. We test whether firm ownership affects carbon dioxide emissions using a provincial data covering the period from 1992 to 2010. In our estimations we separate between short-run and long-run effects using wavelet analysis and a band spectrum regression estimator. Our results offer two important policy implications: First, most of the short-run volatility in emission is due to non-economic factors and policy makers should focus on the long-run where they have a greater possibility to affect emissions growth. Second, over the long-term a larger share of private firms reduces the growth rate in carbon dioxide emissions and continuing structural reforms lifting remaining barriers of private firm production will be crucial to contain or even reduce national emissions.
CO0796: Wavelet-based factor pricing and measurement of macroeconomic risks Presenter: Joanna Bruzda, Nicolaus Copernicus University, Poland
In the paper we suggest the use of analytic wavelets to measure the exposure to different risk factors within multifactor asset pricing models with macroeconomic sources of risks. Our wavelet-based betas are computed as wavelet partial gain coefficients, which incorporate the influence of any lead-lag effects on the values of these risk measures and enable the investor to track changes in systematic risk across wavelet scales. This proposal is illustrated with empirical wavelet APT models for industry portfolios on the US and European stock markets, showing that the modified beta coefficients can substantially change the assessment of macroeconomic risks and demonstrating how they influence the pricing of different risk factors.
CO1004: Long waves in external imbalances, credit growth and asset prices: An historical perspective on global financial crisis Presenter: Marco Gallegati, Polytechnic University of Marche, Italy
The recent global financial crisis has favored a renewed interest in the long run view of macroeconomic history and in Minsky’s contributions on the role the instability of the financial structure for the economy. We use a long-term historical dataset to investigate the timing relationships among long wave patterns in external imbalances, credit growth and asset prices. We document evidence of a recurring sequencing pattern ahead of global crisis periods where credit booms are preceded by growing external imbalances, and financial crisis occur at the low end of the contraction phase of asset price returns. The main implications of our results are the need to pay attention to widening global imbalances and to regulate and reform the financial sector by limiting capital markets movements.
CO1794: The relationship of simple sum and Divisia monetary aggregates with real GDP and inflation: A wavelet analysis for US Presenter: Michael Scharnagl, Deutsche Bundesbank, Germany
Co-authors:Martin Mandler
We apply wavelet analysis to compare the relationship between simple sum and Divisa monetary aggregates with real GDP and CPI inflation for the US using data from 1967 to 2013. Wavelet analysis allows to account for variations in the relationships both across the frequency spectrum and across time. While we find evidence for a weaker comovement of Divisia compared to simple sum monetary aggregates with real GDP the relationship between money growth and inflation is estimated to be much tighter between Divisia monetary aggregates and CPI inflation than for simple sum aggregates, in particular at lower frequencies. Furthermore, for the Divisia indices for broader monetary aggregates (M2, M2M, MZM) we estimate a stable lead before CPI inflation of about four to five years.
CO500 Room Torrington VOLATILITY MODELS Chair: Eduardo Rossi
CO0652: A fractionally co-integrated VAR for predicting equity and index returns Presenter: Marwan Izzeldin, Lancaster University Management School, United Kingdom Co-authors:Xingzhi Yao
A fractionally co-integrated VAR (FCVAR) model is adopted to predict stock/index returns. The model setup allows for the co-integrating combi-nations ofI(d)processes which result in anI(d−b)process, and thus has a wider appeal relative to the simplified case ofI(d)toI(0)currently employed. The empirical analysis is based on the high frequency data for selected stocks and several index indicators (VIX, SPY and S&P 500) observed over the period of (2003-2013). The results show that predictions implied by the FCVAR model are superior to those by the conventional VAR as measured by the R-square constructed from the impulse response functions. The strength of the co-fractional relations between implied-realised volatility as well as the level of return predictability is sensitive to market regimes (pre/post crisis), structural breaks and stock/index degree of activity. The findings highlight the advantages of adopting a fractionally co-integrated framework in predicting returns, especially over long horizons, and the relevance of accounting for regime change and breaks in such a framework.