Past Research

Summary   

Below I provide a brief description of the academic and research trajectory between September 2015 and September 2020.  In particular, during that period I worked as a Research Assistant and Visiting Researcher in various funded research projects at academic institutions such as the University of Cyprus and University College London. Moreover, as a Doctoral Researcher at the University of Southampton, I participated in income generating activities through innovative suggestions for Grant proposals based on related theoretical and empirical studies undertaken during the 'Researcher in Training' stage. Lastly, I provided statistical and econometrics advice for the undertaking of undergraduate dissertations within the fields of economics and econometrics.

On Predictive Accuracy Testing in Big Data Environments 

• Academic Project: Doctoral Researcher Research Project

• Academic Unit: Department of Economics, University of Southampton

• Academic Year: 2019/2020

Main insights

In this research paper we implement statistical tests of equal predictive accuracy (EPA) which allows to evaluate the forecasting power of econometric specifications (forecast evaluation). In particular, we focus on the forecast evaluation of HAR regression models in nested environments and high-dimensional data. Specifically, the high-dimensionality is introduced using cross-sectional volatility measures at different time lags as potential predictors. Under the assumption of regressors sparsity and stationarity we employ penalized functions to assess the performance of the tests to the use of various shrinkage estimators (such as Lasso, Adaptive Lasso, Elastic net), which are employed for model fitting and variable selection purposes. Moreover, the empirical application of this study compares the performance of the EPA testing procedures based on the MSE forecast loss functions. For comparability purposes, these Lasso-based encompassing tests are constructed across nested model specifications, overlapping model specifications and non-overlapping specifications (with respect to the set of candidate predictors). Then, the estimation of the forecasting sequences employs a fixed forecasting scheme (the rolling window of fixed size). Based on the empirical results, there are statistical evidence which indicate that test statistics based on the overlapping specifications have better performance in terms of size distortions (empirical size under the null hypothesis of equal predictive accuracy). From the methodological perspective, since our main research objective is the development of a novel encompassing test for predictive regression models with more general dependence structure, we focus on the nested model environment with forecast errors estimated using an overlapping window. Motivated from the bootstrap unit root and block bootstrap techniques (such as the overlapping block scheme when constructing jackknife-based estimators for near unit root regressions), we conjecture that the proposed overlapping window approach (such as an unequal split-sample within the out-of-sample period is implemented to estimate the sample counterpart of the cross-product moment of the forecast errors of the competing specifications), provides a solution to the variance degeneracy problem; as commonly discussed in the EPA testing literature. 

Recall that for constructing EPA testing procedures, the main idea is to employ recursively estimated forecasts from a restricted regression model (null model) versus an unrestricted regression model (full model). These competing forecast error sequences are based on rolling windows of fixed size, which are also 'chronologically aligned' across competing models. In other words, we assume that without loss of generality, in the case where the forecast sequences from the two competing models are constructed from exact same blocks, the corresponding encompassing test statistics have almost surely equivalent performance (asymptotically equivalent) as in the case where the estimated model variances are obtained from the overlapping window approach. The implementation of the proposed equal predictive accuracy scheme, which includes detailed description of the testing procedure, corresponding asymptotic theory and related Monte Carlo simulation results are presented in the framework of Pitarakis, J-Y. (Econometric Theory, 2023). The research project "On Predictive Accuracy Testing in Big Data Environments", aims to develop automatic dynamic model specification procedures with data-driven shrinkage on the possibly high-dimensional parameter space, to obtain the simplest adequate nested environment regardless of time series persistent properties.  

• Research Tasks: R Coding, HPC usage*, Econometric Modelling, Empirical Application. 

• Research Proposals: 

Proposal 1

Katsouris, C. (April 2020). "Testing for Persistence Change in Nested Predictive Regression Environments". Research Paper Proposal, Department of Economics, University of Southampton. 

     Abstract. This research paper aims to propose a statistical procedure for testing for the presence of predictive accuracy in nested predictive regression systems using information from both the in-sample and the out-of-sample period. Our proposed framework has important implications and is closely related with two streams of literature. First, we aim to derive a framework for a particular form of nested predictive regression environments. The nested property of the proposed predictive regression model structure is used in the sense of 'mixed' persistence (the exact notion of 'heterogeneous persistence' is defined in this proposal). Specifically, in the context of EPA tests, we define the presence of 'heterogeneous persistence' (that holds under the alternative hypothesis) which allows the comparison between a homogeneous persistence predictive regression model (such as predictors belong to the same persistence class as defined in KMS) and the heterogeneous persistence predictive regression model (such as predictors belong to a different persistent class than the persistent class which holds under the null); implying that the homogeneous persistence case is nested in terms of the persistence class of predictors. Second, we propose a testing procedure to examine the null hypothesis of no predictive accuracy of the heterogeneous persistence predictive regression, building on the literature of equal predictive accuracy methodologies. Third, for the empirical application of the paper we use the dataset provided by Hardle et al. (JoE, 2016), which allows to construct and test the proposed nested predictive regression environment.

Proposal 2

Katsouris, C. (Oct 2019). "Forecast Evaluation in Large Cross-Sections of Realized Volatility". Research Paper Proposal, Department of Economics, University of Southampton. 

• Research Outputs: 

1. Katsouris, C. (2020). "Forecast Evaluation in Large Cross-Sections of Realized Volatility". Preprint arXiv:2112.04887.

2. Pitarakis, J. Y. (2020). "A Novel Approach to Predictive Accuracy Testing in Nested Environments". Preprint arXiv:2008.08387.

3. Pitarakis, J. Y. (2021). "Out of Sample Predictability in Predictive Regressions with Many Predictor Candidates". Department of Economics Working paper, University of Southampton. Preprint arXiv:2302.02866.

4. Pitarakis, J. Y. (2025). "Serial-Dependence and Persistence Robust Inference in Predictive Regressions". Preprint arXiv:2502.00475. 

• Limitations for Research Output 2 : 

The framework proposed in the paper of Katsouris, C. (Preprint arXiv:2112.04887, 2020), examines empirically the out-of-sample predictive performance of a cross-sectional regression model with many predictors, based on different shrinkage estimators. Moreover, in the construction of the forecasting scheme we consider the forecast evaluation of the null versus the unrestricted model under the alternative using nested, non-nested and overlapping forecasting environments. However, a limitation of this study is that a formal Monte Carlo simulation study which incorporates changes with respect to an overlapping block approach of the forecast sequences is not implemented. Moreover, we only present a preliminary asymptotic theory on the use of Lasso for estimating model coefficients of a predictive model.

• Further Research : 

A formal asymptotic theory framework which incorporates the forecast evaluation tests is left for future work. In that case, a related framework could be focused on: "Testing Predictive Accuracy in Lasso Predictive Regression Models". Another possible direction is to construct a nested regression environment where the two blocks correspond to a block of 'baseline covariates' (which can be captured via factor structure) and a block of sparse regressors, implying a comparison between a dense model under the null versus a dense plus sparse model under the alternative (such as the Priority Lasso which could be a suitable penalty function for constructing such statistical testing procedures). In this case, a related framework could be focused on: "Testing Predictive Accuracy in Factor-Augmented Predictive Regressions".

• Supervisor: Prof. Jean-Yves Pitarakis (Department of Economics, University of Southampton). PI for the ESRC funded project: "Novel Approaches to Comparing the Predictive Accuracy of Nested Models in Data Rich and Heterogeneous Predictor Environments" (2022-2024).

Ex-Post Related Literature (Impact)

• Corradi, V., Fosten, J., and Gutknecht, D. (2025). "Sparsity Tests for High-Dimensional Time Series Regressions". Available at SSRN 5353643. 

• Morico, A., and Stauskas, O. (2025). "Robust Tests for Factor-Augmented Regressions with an Application to the novel EA-MD Dataset". Preprint arXiv:2504.08455. 

• Margaritella, L. and Stauskas, O. (2024). "New Tests of Equal Forecast Accuracy for Factor-Augmented Regressions with Weaker Loadings". Preprint arXiv:2409.20415. 

• Corradi, V., Fosten, J., and Gutknecht, D. (2024). "Predictive Ability Tests with Possibly Overlapping Models". Journal of Econometrics, 241(1), 105716. 

* We acknowledge the use of the IRIDIS HPC Facility and associated support services at the University of Southampton in the completion of this work. 

On Optimal Portfolio Choice Problem 

• Academic Project: Doctoral Researcher Research Project

• Academic Unit: Department of Economics, University of Southampton

• Academic Year: 2018/2019

Abstract. The traditional portfolio allocation problem has been extensively studied in the risk management and optimisation literature the past decades, providing a way of finding efficient weight allocations on the expected return and risk frontier. However, considering the traditional trade-off as a stand alone dilemma is not a sufficient decision mechanism as the 2008 global financial crisis revealed, which was a period of increased financial connectedness. Building on the optimal portfolio selection problem, we propose a novel optimisation framework within which the network topology is incorporated using a graph-based portfolio selection procedure that relies on an endogenously generated adjacency matrix. 

# portfolio choice, # systemic risk, # network dependence, # iterative eigenvalue gap procedure, # large covariance matrix

• Research Tasks: R Coding, HPC usage*, Econometric Modelling, Simulations-Based Estimation, Empirical Application. 

• Replication Studies:

1. Härdle, W.K., Wang, W., and Yu, L. (2016). "Tenet: Tail-Event Driven Network Risk". Journal of Econometrics, 192(2), 499-513. 

2. Wilms, I., and Croux, C. (2015). "Sparse Cointegration". Working paper, Faculty of Economics and Business, KU Leuven. Preprint arXiv:1501.01250.  

• Research Outputs: 

1. Katsouris, C. (2023). "Robust Estimation in Network Vector Autoregression with Nonstationary Regressors". Preprint arXiv:2401.04050.

2. Katsouris, C. (2023). "Statistical Estimation for Covariance Structures with Tail Estimates using Nodewise Quantile Predictive Regression Models". Preprint arXiv:2305.11282.

3. Katsouris, C. (2021). "Optimal Portfolio Choice and Stock Centrality for Tail Risk Events". Preprint arXiv:2112.12031. 

4. Olmo, J. (2021). "Optimal Portfolio Allocation and Asset Centrality Revisited". Quantitative Finance, 21(9), 1475-1490. 

5. Katsouris, C. (2019). "Literature Review On Volatility Spillovers and the Measurement of Financial Connectedness". PhD Progression Review Paper, Department of Economics, University of Southampton (see Lecture Notes Section).

• Related Research Outputs:

1. Olmo, J. (2023). "A Nonparametric Predictive Regression Model using Partitioning Estimators based on Taylor Expansions". Journal of Time Series Analysis, 44(3), 294-318. 

2. Olmo, J., and Sanso-Navarro, M. (2023). "A Nonparametric Spatial Regression Model using Partitioning Estimators". Econometrics and Statistics. Available at SSRN 3578389.

3. Kapar, B., and Olmo, J. (2022). "A Dynamic Network Regression Model for a Large Cross Section of Units with an Application to Measuring Spillovers between Pollution and Electricity Consumption". Available at SSRN 4072928. 

4. Laborda, R., and Olmo, J. (2021). "Volatility Spillover between Economic Sectors in Financial Crisis Prediction: Evidence Spanning the Great Financial Crisis and Covid-19 Pandemic". Research in International Business and Finance, 57, 101402. 

• Other Related Research Outputs from Southampton Econometrics Group:

1. Mancini, T., Calvo-Pardo, H., and Olmo, J. (2021). "Extremely Randomized Neural Networks for Constructing Prediction Intervals". Neural Networks, 144, 113-128. 

2. Calvo-Pardo, H., Mancini, T., and Olmo, J. (2021). "Granger Causality Detection in High-Dimensional Systems using Feedforward Neural Networks". International Journal of Forecasting, 37(2), 920-940. 

• Supervisor: Prof. Jose Olmo (Department of Economics, University of Southampton)

* We acknowledge the use of the IRIDIS HPC Facility and associated support services at the University of Southampton in the completion of this work. 

On Statistical Validity of Treatment Effect Estimators 

• Academic Project: Research Assistant Research Project, Visiting Researcher (Special Scientist Research)

• Academic Unit: Department of Economics, University College London (UCL)

• Academic Year: 2018/2019

Abstract. Treatment effect estimators are used when evaluating outcomes of randomized control designs, such as health or economic intervention outcomes. In the first phase of this research project we investigate the validity of a permutation test for assessing the statistical significance of the treatment effect estimator under the experimental conditions of baseline imbalance and attrition using finite-sample Monte Carlo simulation experiments. In the second phase of the project, we examine the implementation of post-selection inference methods in a many covariates setting for a health survey study.

# treatment effect estimation, # permutation test

• Research Tasks: Monte Carlo Simulations, Stata Coding, HPC usage, Econometric Modelling, Empirical Application. 

• Research Outputs: 

1. Monte Carlo Permutation Test Validation Study (Preprint). 

2. "Post-Model Selection Inference for a Child Development Survey Study" (Working paper, 2019).

3. "Testing Statistical Significance in Linear Models with Many Controls under Hidden Confounding" (Working paper, June 2023). 

• Supervisors: Prof. Gabriella Conti (UCL) and Dr. Stavros Poupakis (UCL).

• Research Participants: Christis Katsouris (Southampton), Giacomo Mason (Economics Ph.D. Candidate, UCL), Jiaqi Li (Postgraduate student, UCL).

• Funding: ERC Grant "DynaHEALTH" (PI Prof. Gabriella Conti, Department of Economics UCL).

* We acknowledge the use of the HPC Facility and associated support services at the University College London in the completion of this work.