CHEVILLON Guillaume

In this thesis, we compare IMS, DMS and PLS forecasts at multiple horizons, focusing on the combinatorial properties of PLS. We build on an interesting paper by Franses & Legerstee (2010), which suggests how the so-called partial least squares (PLS) method can be considered, in the context of multi-step forecasting, as an intermediate technique between IMS and DMS. In fact, rather than an "intermediate", we like to think of PLS as a form of combination of IMS and DMS.This thesis consists of four chapters.In Chapter 1, we provide a review of the literature that serves as background for the following chapters. In Chapter 2, we explore the functionality of PLS as a combination of IMS and DMS. We study the properties of PLS using an algorithm suggested by Garthwaite (1994).We investigate the relationship between IMS, DMS and PLS and compare the accuracy of their predictions at several horizons. We analyze the combinatorial properties of PLS in multistage forecasting using a simple AR model (2). To compare forecasting performance, we evaluate their asymptotic properties under well-specified and misspecified models. We confirm our analytical study through extensive simulations of the relative forecast accuracy of different multistage forecasting techniques. Through these simulations, we support the asymptotic analysis and investigate the conditions that make PLS better than IMS or DMS.In Chapter 3, we conduct an empirical study of multi-step forecasting based on univariate AR models and focus on the 121 monthly macroeconomic time series in the U.S. We provide an empirical analysis to determine the circumstances that make PLS preferable to IMS or DMS. For easier comparison with the literature, we follow Marcellino et al. (2006) and McCracken & McGillicuddy (2019) in many respects. In addition, we extend their results in some directions, such as path prediction evaluation, alternative measurement techniques, and different subsamples.We explore the benefits in relation to the persistence of the series measured by the degree of fractional integration.Through this empirical analysis, we reconfirm the results of previous studies and discover several new facts: (i) the relative advantages of PLS over IMS tend to disappear as the forecast horizon expands. (ii) PLS is generally better when the model uses short lags. and (iii) PLS performs better than IMS when the data undergo periods of high volatility.The final chapter extends Chapter 3 to multivariate models. We compare a brief analytical study of the rationale for PLS and then empirically compare the forecasting performance of IMS, IMS, and DMS in the context of bivariate forecasting models. For each forecasting model, we generate and evaluate forecasts over a single horizon and over trajectories (ranges of horizons). Our results confirm those of the univariate models: PLS is favored in the short run, but the crucial issue is data persistence. In this respect, the data for the nominal prices, wages and money group show a form of persistence that does not clearly follow an I (1) or I (2) trend and produces much better PLS performance. Overall, we also find that PLS is generically preferred to DMS, so it should be an alternative for the practitioner whenever direct forecasting techniques can be considered.

This paper shows that a large dimensional vector autoregressive model (VAR) of finite order can generate fractional integration in the marginalized univariate series. We derive high-level assumptions under which the final equation representation of a VAR(1) leads to univariate fractional white noises and verify the validity of these assumptions for two specific models.

No summary available.

This paper shows that a large dimensional vector autoregressive model (VAR) of finite order can generate fractional integration in the marginalized univariate series. We derive high-level assumptions under which the final equation representation of a VAR(1) leads to univariate fractional white noises and verify the validity of these assumptions for two specific models.

This paper studies the properties of multi-step projections, and forecasts that are obtained using either iterated or direct methods. The models considered are local asymptotic: they allow for a near unit root and a local to zero drift. We treat short, intermediate and long term forecasting by considering the horizon in relation to the observable sample size. We show the implication of our results for models of predictive regressions used in the financial literature. We show here that direct projection methods at intermediate and long horizons are robust to the potential misspecification of the serial correlation of the regression errors. We therefore recommend, for better global power in predictive regressions, a combination of test statistics with and without autocorrelation correction.

We study learning dynamics in a prototypical representative-agent forwardlooking model in which agents’ beliefs are updated using linear learning algorithms. We show that learning in this model can generate long memory endogenously, without any persistence in the exogenous shocks, depending on the weights agents place on past observations when they update their beliefs, and on the magnitude of the feedback from expectations to the endogenous variable. This is distinctly different from the case of rational expectations, where the memory of the endogenous variable is determined exogenously.

Long-run restrictions are a very popular method for identifying structural vector autoregressions, but they suffer from weak identification when the data is very persistent, i.e., when the highest autoregressive roots are near unity. Near unit roots introduce additional nuisance parameters and make standard weak-instrument-robust methods of inference inapplicable. We develop a method of inference that is robust to both weak identification and strong persistence. The method is based on a combination of the Anderson-Rubin test with instruments derived by filtering potentially non-stationary variables to make them near stationary. We apply our method to obtain robust confidence bands on impulse responses in two leading applications in the literature.

This paper studies the properties of iterated and direct multistep forecasting techniques in the presence of in-sample location shifts (breaks in the mean). It also considers the interactions of these techniques with multistep intercept corrections that are designed to exhibit robustness to such shifts. In a local-asymptotic parameterization of the probability of breaks, we provide analytical expressions for forecast biases and mean-square forecast errors. We also provide simulations which show that breaks provide a rationale for using methods other than iterated multistep techniques. In particular, we study the relationships between the relative accuracy of the methods and the forecast horizon, the sample size and the timing of the shifts. We show that direct multistep forecasting provides forecasts that are relatively robust to breaks, and that its benefits increase with the forecast horizon. In an empirical application, we revisit an oft-used dataset of G7 macroeconomic series and corroborate our theoretical results.

This thesis consists of four essays on finance and the real economy. Chapter 1 studies the effect of banking crises on the composition of investment. It builds a partial equilibrium growth model with a banking sector and two types of investment projects: a safe, low return technology and an innovative, high productivity one. Investments in innovation are risky since they are subject to a liquidity cost which entrepreneurs cover by borrowing from the banking sector. When bank creditors are sufficiently pessimistic about the aggregate liquidity needs of the real sector, they will run on the bank and cause a credit freeze. This leads banks to tighten credit supply after the crisis, which decreases disproportionately investment in innovation and slows down economic growth. An empirical investigation, employing industry-level data on R&D investment around 13 recent banking crises, confirms this hypothesis. Industries that depend more on external finance, in more bank-based economies, invest disproportionately less in R&D following episodes of banking distress. These industries also have a relatively lower share of R&D in total investment, suggesting a shift in the composition of investment after the crisis. Such differential effects across sectors imply that the drop in R&D spending is, at least partially, the result of the contraction in credit supply.Chapter 2 studies the impact of coordination frictions in financial markets on the cost of capital. In the model, a financial intermediary seeks to raise funds to finance a risky capital-intensive project. Capital is owned by a large number of small investors, who observe noisy signals about the project's implementation cost. Employing a global games equilibrium refinement, we characterize a unique threshold equilibrium of the coordination game between investors. We then show that the relationship between the probability of success of the project and the rate of return on capital is non-monotonic. There exists a socially optimal price of capital, which maximizes the probability that the project is profitable. However, fee-maximizing intermediaries will generally set an interest rate that is higher than the socially optimal rate. The model best characterizes project finance investments funded through the bond market.Chapter 3 proposes a laboratory experiment to study the impact of partial deposit insurance schemes on the risk of deposit withdrawals. In the experiment, depositors decide whether to withdraw or leave their money in a bank, triggering a default when too many participants choose to withdraw. When a bank run occurs, the amount of wealth each depositor can recover depends on the number of withdrawals and a deposit insurance fund whose size cannot cover in full all depositors. We consider two treatments: (i) a perfect information case when depositors know the size of the insurance fund and (ii) a heterogeneous information setting when they only observe noisy signals about its size. Our results show that uncertainty about the level of deposit coverage exerts a significant impact on the propensity to run. The frequency of runs is relatively high in both treatments. A majority of subjects follow a threshold strategy consistent with a risk-dominant equilibrium selection. Finally, the last chapter re-examines the empirical relationship between financial and economic development while (i) taking into account their dynamics and (ii) differentiating between stock market and banking sector development. We study the cointegration and causality between finance and growth for 22 advanced economies. Our time series analysis suggests that the evidence in support of a finance-led growth is weak once we take into account the dynamics of financial and economic development. We show that, causality patterns depend on whether countries' financial development stems from the stock market or the banking sector.

This paper shows that large dimensional vector autoregressive (VAR) models of finite order can generate long memory in the marginalized univariate series. We derive high-level assumptions under which the final equation representation of a VAR(1) leads to univariate fractional white noises and verify the validity of these assumptions for two specific models. We consider the implications of our findings for the variances of asset returns where the so-called golden-rule of realized variances states that they tend always to exhibit fractional integration of a degree close to 0:4.

This paper shows that large dimensional vector autoregressive (VAR) models of fi nite order can generate long memory in the marginalized univariate series. We derive high-level assumptions under which the fi nal equation representation of a VAR(1) leads to univariate fractional white noises and verify the validity of these assumptions for two speci fic models. We consider the implications of our findings for the variances of asset returns where the so-called golden-rule of realized variances states that they tend always to exhibit fractional integration of a degree close to 0:4.

No summary available.

No summary available.

This thesis studies several aspects of the dynamics of asset returns. The first three chapters focus on price formation in the art market. The first chapter establishes that prices can temporarily, and partially predictably, deviate from fundamental value. This article was published in Economics Letters (Volume 122, Issue 3, pp. 432-434) and was written with Christophe Spaenjers and Luc Renneboog. Chapter 2 studies the speed of information transmission in aggregate art market prices. Chapter 3 analyzes the correlation between price and volume and supports evidence consistent with a bubble hypothesis. It was written with Luc Renneboog. Chapter 4 focuses on empirical modeling of the predictability of stock market indices in fifteen industrialized countries. It proposes to combine the information given by each country in order to improve the predictive power.

Standard tests for the rank of cointegration of a vector autoregressive process present distributions that are affected by the presence of deterministic trends. We consider the recent approach of Demetrescu et al. (2009) who recommend testing a composite null. We assess this methodology in the presence of trends (linear or broken) whose magnitude is small enough not to be detectable at conventional significance levels. We model them using local asymptotics and derive the properties of the test statistics. We show that whether the trend is orthogonal to the cointegrating vector has a major impact on the distributions but that the test combination approach remains valid. We apply of the methodology to the study of cointegration properties between global temperatures and the radiative forcing of human gas emissions. We find new evidence of Granger Causality.

We consider a prototypical representative-agent forward-looking model, and study the low frequency variability of the data when the agent's beliefs about the model are updated through linear learning algorithms. We find that learning in this context can generate strong persistence. The degree of persistence depends on the weights agents place on past observations when they update their beliefs, and on the magnitude of the feedback from expectations to the endogenous variable. When the learning algorithm is recursive least squares, long memory arises when the coefficient on expectations is sufficiently large. In algorithms with discounting, long memory provides a very good approximation to the low-frequency variability of the data. Hence long memory arises endogenously, due to the self-referential nature of the model, without any persistence in the exogenous shocks. This is distinctly different from the case of rational expectations, where the memory of the endogenous variable is determined exogenously. Finally, this property of learning is used to shed light on some well-known empirical puzzles.

This paper proposes a Near Explosive Random-Coefficient autoregressive model for asset pricing which accommodates both the fundamental asset value and the recurrent presence of autonomous deviations or bubbles. Such a process can be stationary with or without fat tails, unit-root nonstationary or exhibit temporary exponential growth. We develop the asymptotic theory to analyze ordinary least-squares (OLS) estimation. One important theoretical observation is that the estimator distribution in the random coefficient model is qualitatively different from its distribution in the equivalent fixed coefficient model. We conduct recursive and full-sample inference by inverting the asymptotic distribution of the OLS test statistic, a common procedure in the presence of localizing parameters. This methodology allows to detect the presence of bubbles and establish probability statements on their apparition and devolution. We apply our methods to the study of the dynamics of the Case-Shiller index of U.S. house prices. Focusing in particular on the change in the price level, we provide an early detection device for turning points of booms and bust of the housing market.

The following three chapters examine: 1) whether Southeast Asian real exchange rates are nonlinear, 2) Bayesian inference on nonlinear time series model with applications to the real exchange rate, and 3) cyclicality and rebound effect in the stock market.Since the late 1990s, theoretical and empirical analyses devoted to the real exchange rate suggest that the dynamics could be well estimated by nonlinear models. The first chapter examines this possibility using monthly ASEAN-5 data, and extends existing research in two directions. First, we use recently developed unit root tests which will allow for more flexibility in stationary nonlinear models as an alternative to the commonly used SETAR or ESTAR model. Second, although different nonlinear models survive mis-specification tests, a Monte Carlo experiment using generalized impulse response functions is used to compare their relative adequacy. Our results i) support the nonlinear mean-reversion hypothesis, and thus purchasing power parity, in half of the cases and ii) indicate MRLSTAR and ESTAR as the most likely processes generating real exchange rates.The second chapter analyzes ACR model. We propose a full Bayesian approach to inference and particular attention is paid to the parameters of the threshold variables. We discuss the choice of a priori distributions and propose a Markov chain Monte Carlo algorithm to estimate the parameters and latent variables. A simulation study and application to real exchange rate data illustrate the analysis.The third chapter explores that different forms of overlaps in financial markets can present in a Markov Switching model. It builds on the rebound effects first analyzed by Kim, Morley and Piger [2005] in the business cycle and generalized by Bec, Bouabdallah and Ferrara [2011] to allow for a more flexible rebound type.Our results i) show that the rebound effect is statistically significant and important in all cases, but Germany where the evidence is less clear and ii) the permanent negative impact of bear markets on the index is significantly reduced when the rebound is explicitly taken into account.