nep-ecm New Economics Papers
on Econometrics
Issue of 2021‒09‒13
thirteen papers chosen by
Sune Karlsson
Örebro universitet

  1. Variable Selection for Causal Inference via Outcome-Adaptive Random Forest By Daniel Jacob
  2. Ubiquitous multimodality in mixed causal-noncausal processes. By Kindop, Igor
  3. Inferential Theory for Generalized Dynamic Factor Models By Matteo Barigozzi; Marc Hallin; Matteo Luciani; Paolo Zaffaroni
  4. Multi Anchor Point Shrinkage for the Sample Covariance Matrix (Extended Version) By Hubeyb Gurdogan; Alec Kercheval
  5. A generalized bootstrap procedure of the standard error and confidence interval estimation for inverse probability of treatment weighting By Tenglong Li; Jordan Lawson
  6. Approximate Factor Models with Weaker Loadings By Jushan Bai; Serena Ng
  7. Some Impossibility Results for Inference With Cluster Dependence with Large Clusters By Denis Kojevnikov; Kyungchul Song
  8. On a quantile autoregressive conditional duration model applied to high-frequency financial data By Helton Saulo; Narayanaswamy Balakrishnan; Roberto Vila
  9. Estimation of the Financial Cycle with a Rank-Reduced Multivariate State-Space Model By Rob Luginbuhl
  10. Forecasting Dynamic Term Structure Models with Autoencoders By Castro-Iragorri, C; Ramírez, J
  11. On the estimation of discrete choice models to capture irrational customer behaviors By Sanjay Dominik Jena; Andrea Lodi; Claudio Sole
  12. Creating Powerful and Interpretable Models with Regression Networks By Lachlan O'Neill; Simon D Angus; Satya Borgohain; Nader Chmait; David Dowe
  13. Iterated and exponentially weighted moving principal component analysis By Paul Bilokon; David Finkelstein

  1. By: Daniel Jacob
    Abstract: Estimating a causal effect from observational data can be biased if we do not control for self-selection. This selection is based on confounding variables that affect the treatment assignment and the outcome. Propensity score methods aim to correct for confounding. However, not all covariates are confounders. We propose the outcome-adaptive random forest (OARF) that only includes desirable variables for estimating the propensity score to decrease bias and variance. Our approach works in high-dimensional datasets and if the outcome and propensity score model are non-linear and potentially complicated. The OARF excludes covariates that are not associated with the outcome, even in the presence of a large number of spurious variables. Simulation results suggest that the OARF produces unbiased estimates, has a smaller variance and is superior in variable selection compared to other approaches. The results from two empirical examples, the effect of right heart catheterization on mortality and the effect of maternal smoking during pregnancy on birth weight, show comparable treatment effects to previous findings but tighter confidence intervals and more plausible selected variables.
    Date: 2021–09
  2. By: Kindop, Igor
    Abstract: According to the literature, the bimodality of estimates in mixed causal–non-causal autoregressive processes is due to unlucky starting values and happens only ocassionally. This paper shows that a unique and convergent solution is not always the case for models of this class. Instead, the likelihood function is not convex leading to the multimodality of estimated parameters. It can be attributed to the magnitude and sign of the autoregressive coefficients. Simultaneously, the number of local modes grows with the number of autoregressive parameters in the model. This multimodality depends on the parameters of the process and the chosen error distribution. We have to apply grid search methods to extract candidate solutions. The independence of residuals is a necessary hypothesis for the proper identification of the processes. A simple AIC criterion helps to select an independent model. Finally, I sketch a roadmap on estimating mixed causal-noncausal autoregressive models and illustrate the approach with Brent spot oil price returns.
    Keywords: non-causal model, non-convex likelihood, non-Gaussian, nonfundamentalness, multimodality.
    JEL: C13 C22 C51 C52 C53 E37
    Date: 2021–07–09
  3. By: Matteo Barigozzi; Marc Hallin; Matteo Luciani; Paolo Zaffaroni
    Abstract: We provide the asymptotic distributional theory for the so-called General or Generalized Dynamic Factor Model (GDFM), laying the foundations for an inferential approach in the GDFM analysis of high-dimensional time series. Our results are exploiting the duality between common shocksand dynamic loadings under a random cross-section approach to derive the asymptotic distribution of a class of estimators for common shocks, dynamic loadings, common components, and impulse response functions. An empirical application aimed at the construction of a “core” inflation indicator for the U.S. economy is presented, empirically demonstrating the superiority of the GDFM-based indicator over the most commonly adopted approaches, outperforming, in particular, the one based on Principal Components.
    Keywords: High-dimensional time series, Generalized Dynamic Factor Models, One-sided representations of dynamic factor models, Asymptotic distribution, Confidence intervals
    Date: 2021–08
  4. By: Hubeyb Gurdogan; Alec Kercheval
    Abstract: Portfolio managers faced with limited sample sizes must use factor models to estimate the covariance matrix of a high-dimensional returns vector. For the simplest one-factor market model, success rests on the quality of the estimated leading eigenvector "beta". When only the returns themselves are observed, the practitioner has available the "PCA" estimate equal to the leading eigenvector of the sample covariance matrix. This estimator performs poorly in various ways. To address this problem in the high-dimension, limited sample size asymptotic regime and in the context of estimating the minimum variance portfolio, Goldberg, Papanicolau, and Shkolnik developed a shrinkage method (the "GPS estimator") that improves the PCA estimator of beta by shrinking it toward a constant target unit vector. In this paper we continue their work to develop a more general framework of shrinkage targets that allows the practitioner to make use of further information to improve the estimator. Examples include sector separation of stock betas, and recent information from prior estimates. We prove some precise statements and illustrate the resulting improvements over the GPS estimator with some numerical experiments.
    Date: 2021–08
  5. By: Tenglong Li; Jordan Lawson
    Abstract: The inverse probability of treatment weighting (IPTW) approach is commonly used in propensity score analysis to infer causal effects in regression models. Due to oversized IPTW weights and errors associated with propensity score estimation, the IPTW approach can underestimate the standard error of causal effect. To remediate this, bootstrap standard errors have been recommended to replace the IPTW standard error, but the ordinary bootstrap (OB) procedure might still result in underestimation of the standard error because of its inefficient sampling algorithm and un-stabilized weights. In this paper, we develop a generalized bootstrap (GB) procedure for estimating the standard error of the IPTW approach. Compared with the OB procedure, the GB procedure has much lower risk of underestimating the standard error and is more efficient for both point and standard error estimates. The GB procedure also has smaller risk of standard error underestimation than the ordinary bootstrap procedure with trimmed weights, with comparable efficiencies. We demonstrate the effectiveness of the GB procedure via a simulation study and a dataset from the National Educational Longitudinal Study-1988 (NELS-88).
    Date: 2021–08
  6. By: Jushan Bai; Serena Ng
    Abstract: Pervasive cross-section dependence is increasingly recognized as an appropriate characteristic of economic data and the approximate factor model provides a useful framework for analysis. Assuming a strong factor structure, early work established convergence of the principal component estimates of the factors and loadings to a rotation matrix. This paper shows that the estimates are still consistent and asymptotically normal for a broad range of weaker factor loadings, albeit at slower rates and under additional assumptions on the sample size. Standard inference procedures can be used except in the case of extremely weak loadings which has encouraging implications for empirical work. The simplified proofs are of independent interest.
    Date: 2021–09
  7. By: Denis Kojevnikov; Kyungchul Song
    Abstract: There have appeared in the literature various approaches of inference for models with cluster dependence with few clusters, where the dependence structure within each cluster is unknown. These approaches are all different from the "standard" approach of asymptotic inference based on an asymptotically pivotal test. One may wonder whether it is possible to develop a standard asymptotic inference in this situation. To answer this question, we focus on a Gaussian experiment, and present a necessary and sufficient condition for the cluster structure that the long run variance is consistently estimable. Our result implies that when there is at least one large cluster, the long run variance is not consistently estimable, and hence, the standard approach of inference based on an asymptotically pivotal test is not possible. This impossibility result extends to other models that contain a Gaussian experiment as a special case. As a second focus, we investigate the consistent discrimination of the common mean from a sample with cluster dependence. We show that when the observations consist of large clusters, it is necessary for the consistent discrimination of the mean that the sample has at least two large clusters. This means that if one does not know the dependence structure at all, it is not possible to consistently discriminate the mean.
    Date: 2021–09
  8. By: Helton Saulo; Narayanaswamy Balakrishnan; Roberto Vila
    Abstract: Autoregressive conditional duration (ACD) models are primarily used to deal with data arising from times between two successive events. These models are usually specified in terms of a time-varying conditional mean or median duration. In this paper, we relax this assumption and consider a conditional quantile approach to facilitate the modeling of different percentiles. The proposed ACD quantile model is based on a skewed version of Birnbaum-Saunders distribution, which provides better fitting of the tails than the traditional Birnbaum-Saunders distribution, in addition to advancing the implementation of an expectation conditional maximization (ECM) algorithm. A Monte Carlo simulation study is performed to assess the behavior of the model as well as the parameter estimation method and to evaluate a form of residual. A real financial transaction data set is finally analyzed to illustrate the proposed approach.
    Date: 2021–09
  9. By: Rob Luginbuhl (CPB Netherlands Bureau for Economic Policy Analysis)
    Abstract: We propose a model-based method to estimate a unique financial cycle based on a rank-restricted multivariate state-space model. This permits us to use mixed-frequency data, allowing for longer sample periods. In our model the financial cycle dynamics are captured by an unobserved trigonometric cycle component. We identify a single financial cycle from the multiple time series by imposing rank reduction on this cycle component. The rank reduction can be justified based on a principal components argument. The model also includes unobserved components to capture the business cycle, time-varying seasonality, trends, and growth rates in the data. In this way we can control for these effects when estimating the financial cycle. We apply our model to US and Dutch data and conclude that a bivariate model of credit and house prices is sufficient to estimate the financial cycle.
    JEL: E5 F3 G15 G01
    Date: 2020–02
  10. By: Castro-Iragorri, C; Ramírez, J
    Abstract: Principal components analysis (PCA) is a statistical approach to build factor models in finance. PCA is also a particular case of a type of neural network known as an autoencoder. Recently, autoencoders have been successfully applied in financial applications using factor models, Gu et al. (2020), Heaton and Polson (2017). We study the relationship between autoencoders and dynamic term structure models; furthermore we propose different approaches for forecasting. We compare the forecasting accuracy of dynamic factor models based on autoencoders, classical models in term structure modelling proposed in Diebold and Li (2006) and neural network-based approaches for time series forecasting. Empirically, we test the forecasting performance of autoencoders using the U.S. yield curve data in the last 35 years. Preliminary results indicate that a hybrid approach using autoencoders and vector autoregressions framed as a dynamic term structure model provides an accurate forecast that is consistent throughout the sample. This hybrid approach overcomes in-sample overfitting and structural changes in the data.
    Keywords: autoencoders, factor models, principal components, recurrentneural networks
    JEL: C45 C53 C58
    Date: 2021–07–29
  11. By: Sanjay Dominik Jena; Andrea Lodi; Claudio Sole
    Abstract: The Random Utility Maximization model is by far the most adopted framework to estimate consumer choice behavior. However, behavioral economics has provided strong empirical evidence of irrational choice behavior, such as halo effects, that are incompatible with this framework. Models belonging to the Random Utility Maximization family may therefore not accurately capture such irrational behavior. Hence, more general choice models, overcoming such limitations, have been proposed. However, the flexibility of such models comes at the price of increased risk of overfitting. As such, estimating such models remains a challenge. In this work, we propose an estimation method for the recently proposed Generalized Stochastic Preference choice model, which subsumes the family of Random Utility Maximization models and is capable of capturing halo effects. Specifically, we show how to use partially-ranked preferences to efficiently model rational and irrational customer types from transaction data. Our estimation procedure is based on column generation, where relevant customer types are efficiently extracted by expanding a tree-like data structure containing the customer behaviors. Further, we propose a new dominance rule among customer types whose effect is to prioritize low orders of interactions among products. An extensive set of experiments assesses the predictive accuracy of the proposed approach. Our results show that accounting for irrational preferences can boost predictive accuracy by 12.5% on average, when tested on a real-world dataset from a large chain of grocery and drug stores.
    Date: 2021–09
  12. By: Lachlan O'Neill (Faculty of Information Technology, Monash University); Simon D Angus (Dept. of Economics & SoDa Laboratories, Monash Business School, Monash University); Satya Borgohain (SoDa Laboratories, Monash Business School, Monash University); Nader Chmait (Faculty of Information Technology, Monash University); David Dowe (Faculty of Information Technology, Monash University)
    Abstract: As the discipline has evolved, research in machine learning has been focused more and more on creating more powerful neural networks, without regard for the interpretability of these networks. Such “black-box models†yield state-of-the-art results, but we cannot understand why they make a particular decision or prediction. Sometimes this is acceptable, but often it is not. We propose a novel architecture, Regression Networks, which combines the power of neural networks with the understandability of regression analysis. While some methods for combining these exist in the literature, our architecture generalizes these approaches by taking interactions into account, offering the power of a dense neural network without forsaking interpretability. We demonstrate that the models exceed the state-of-the-art performance of interpretable models on several benchmark datasets, matching the power of a dense neural network. Finally, we discuss how these techniques can be generalized to other neural architectures, such as convolutional and recurrent neural networks.
    Keywords: machine learning, policy evaluation, neural networks, regression, classification
    JEL: C45 C14 C52
    Date: 2021–09
  13. By: Paul Bilokon; David Finkelstein
    Abstract: The principal component analysis (PCA) is a staple statistical and unsupervised machine learning technique in finance. The application of PCA in a financial setting is associated with several technical difficulties, such as numerical instability and nonstationarity. We attempt to resolve them by proposing two new variants of PCA: an iterated principal component analysis (IPCA) and an exponentially weighted moving principal component analysis (EWMPCA). Both variants rely on the Ogita-Aishima iteration as a crucial step.
    Date: 2021–08

This nep-ecm issue is ©2021 by Sune Karlsson. It is provided as is without any express or implied warranty. It may be freely redistributed in whole or in part for any purpose. If distributed in part, please include this notice.
General information on the NEP project can be found at For comments please write to the director of NEP, Marco Novarese at <>. Put “NEP” in the subject, otherwise your mail may be rejected.
NEP’s infrastructure is sponsored by the School of Economics and Finance of Massey University in New Zealand.