CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
Shrinkage estimation methods for mixed data sampling regression and heterogeneous autoregressive models
Jönköping University, Jönköping International Business School, JIBS, Statistics.ORCID iD: 0000-0003-4793-9683
2024 (English)Doctoral thesis, comprehensive summary (Other academic)
Abstract [en]

This thesis consists of four research articles. The first two articles assess the effectiveness of various shrinkage estimation methods within mixed data sampling (MIDAS) regression models and find that our proposed methods have superior performance compared to existing models. The third article extends MIDAS models to encompass count data, and the fourth article evaluates the performance of a specific proposed shrinkage method in forecasting stock price volatility.

In the first article, which focuses on MIDAS regression in a nonparametric way, two-parameter nonparametric shrinkage estimation methods are developed to estimate the MIDAS regression parameters. The proposed methodology is compared with one-parameter nonparametric and parametric MIDAS regression, both theoretically via simulation and practically in terms of forecasting U.S. inflation rates. The proposed two-parameter estimator outperforms the one-parameter estimator and other comparative methods, both theoretically and empirically.

In the second article, the two-dimensional panel data regression model is extended to a multidimensional context for mixed-frequency data. We use the least absolute shrinkage and selection operator (LASSO), sparse group (sg)-LASSO, and elastic net unrestricted MIDAS (U-MIDAS) for estimation. The theoretical properties of the extended models are evaluated using Monte Carlo simulations. The proposed model is empirically applied to now cast three-dimensional home ownership vacancy rates across states, metropolitan statistical areas (MSAs), and time in the U.S. Finally, we compare the predictive performance of this extended model with the traditional three-dimensional panel data regression model. The extended model demonstrates superior performance over traditional multidimensional methods, both theoretically and empirically.

The third article introduces a generalized Poisson regression model for count time series data, applied within a MIDAS framework. The new MIDAS Poisson regression model (MIDAS-PRM) is used to forecast the monthly dengue counts from high-frequency environmental parameters and Google Trends data.

Forecasts are generated using a rolling window forecast scheme and forecast combinations. We conclude that the proposed MIDAS-PRM significantly enhances predictive performance compared to the standard time series PRM and other benchmark time series models.

The fourth article proposes two-parameter ridge shrinkage estimation methods to estimate the realized volatility (RV) of the heterogeneous autoregressive (HAR) model. The proposed estimator, which is notable for its orthogonality properties, is employed to forecast the RV of stock prices. The proposed estimator is evaluated through simulations and empirical applications, demonstrating superior performance both theoretically and empirically compared to traditional methods.

Abstract [sv]

Denna doktorsavhandling består av fyra forskningsartiklar. De två första artiklarna utvärderar effektiviteten för olika krympningsmetoder inom mixed data sampling (MIDAS). De föreslagna modellerna presterar bättre än modeller från tidigare forskning. Den tredje artikeln utvidgar MIDAS-modellerna till att omfatta så kallade ”count variables”, medan den fjärde artikeln utvärderar vår föreslagna krympningsmetod för att prognostisera aktiekursvolatilitet. I den första artikeln utvecklas icke-parametriska krympningsmetoder med två parametrar för att estimera MIDAS-modellen.

Den föreslagna metodologin jämförs med konventionella en-parameterbaserade icke-parametriska och parametriska MIDAS-regressionsmodeller. Detta görs både genom teoretiska simuleringar och praktiska tillämpningar vid prognostisering av amerikanska inflationsdata. Den föreslagna två-parameterestimatorn överträffar en-parameterestimatorn och andra traditionella metoder, både teoretiskt och empiriskt.

I den andra artikeln utökar vi den tvådimensionella paneldatamodellen till en multidimensionell kontext för data med blandade datafrekvenser. Vi använder least absolute shrinkage and selection operator (LASSO), sparse group (sg)-LASSO och elastic net unrestricted MIDAS (U-MIDAS) vid estimationen. Dessutom utvärderar vi de teoretiska egenskaperna hos de utökade modellerna genom Monte Carlo-simulering. Sedan använder vi den föreslagna modellen för att utföra tredimensionella nulägesprognoser (nowcasting) för vakanser av äganderätter för olika stater, storstadsområden (metropolitan statistical areas) och över tid i USA. Slutligen jämför vi prediktionsförmågan hos denna utökade modell med den traditionella tredimensionella paneldatamodellen. Resultaten påvisar att den utökade modellen presterar bättre än de traditionella multidimensionella metoderna, både från ett teoretiskt och empiriskt perspektiv.

Den tredje artikeln introducerar en generaliserad Poisson-regression för s.k. count data” som tillämpas inom en MIDAS-kontext. Den nya MIDAS Poisson-modellen (MIDAS-PRM) används för att prognostisera månatliga fall av denguefeber baserat på högfrekventa miljöparametrar och Google Trends-data.

Prognoserna genereras med hjälp av ett rullande fönster och kombinationer av olika prognosmodeller. Artikeln konkluderar att den föreslagna MIDAS-PRM förbättrar de prediktiva egenskaperna jämfört med traditionell tidsserie-PRM och andra benchmarkmodeller.

I den fjärde artikeln föreslås ridge-krympningsmetoder med två parametrar i syfte att estimera den realiserade volatiliteten (RV) genom en heterogen autoregressiv (HAR) modell. Den föreslagna estimatorn, känd för sina goda ortogonalitetsegenskaper, används för att prognostisera realiserad volatilitet för aktiekurser. Både empiriskt och teoretiskt presterar den föreslagna estimatorn bättre än de alternativa estimatorerna, vilket påvisas genom simuleringar och empiriska tillämpningar.

Place, publisher, year, edition, pages
Jönköping: Jönköping University, Jönköping International Business School , 2024. , p. 29
Series
JIBS Dissertation Series, ISSN 1403-0470 ; 165
National Category
Probability Theory and Statistics
Identifiers
URN: urn:nbn:se:hj:diva-65985ISBN: 978-91-7914-044-1 (print)ISBN: 978-91-7914-045-8 (electronic)OAI: oai:DiVA.org:hj-65985DiVA, id: diva2:1890050
Public defence
2024-09-13, B1014, Jönköping International Business School, Jönköping, 10:00 (English)
Opponent
Supervisors
Available from: 2024-08-19 Created: 2024-08-19 Last updated: 2024-08-19Bibliographically approved
List of papers
1. Improved Breitung and Roling estimator for mixed-frequency models with application to forecasting inflation rates
Open this publication in new window or tab >>Improved Breitung and Roling estimator for mixed-frequency models with application to forecasting inflation rates
2024 (English)In: Statistical papers, ISSN 0932-5026, E-ISSN 1613-9798, Vol. 65, p. 3303-3325Article in journal (Refereed) Published
Abstract [en]

Instead of applying the commonly used parametric Almon or Beta lag distribution of MIDAS, Breitung and Roling (J Forecast 34:588–603, 2015) suggested a nonparametric smoothed least-squares shrinkage estimator (henceforth SLS1) for estimating mixed-frequency models. This SLS1 approach ensures a flexible smooth trending lag distribution. However, even if the biasing parameter in SLS1 solves the overparameterization problem, the cost is a decreased goodness-of-fit. Therefore, we suggest a modification of this shrinkage regression into a two-parameter smoothed least-squares estimator (SLS2). This estimator solves the overparameterization problem, and it has superior properties since it ensures that the orthogonality assumption between residuals and the predicted dependent variable holds, which leads to an increased goodness-of-fit. Our theoretical comparisons, supported by simulations, demonstrate that the increase in goodness-of-fit of the proposed two-parameter estimator also leads to a decrease in the mean square error of SLS2, compared to that of SLS1 . Empirical results, where the inflation rate is forecasted based on the oil returns, demonstrate that our proposed SLS2 estimator for mixed-frequency models provides better estimates in terms of decreased MSE and improved R2, which in turn leads to better forecasts.

Place, publisher, year, edition, pages
Springer, 2024
Keywords
Forecast, Inflation, MIDAS, Oil returns, Shrinkage estimator, Smooth least squares estimator
National Category
Probability Theory and Statistics
Identifiers
urn:nbn:se:hj:diva-63378 (URN)10.1007/s00362-023-01520-2 (DOI)001135846800001 ()2-s2.0-85181444604 (Scopus ID)HOA;intsam;928378 (Local ID)HOA;intsam;928378 (Archive number)HOA;intsam;928378 (OAI)
Available from: 2024-01-16 Created: 2024-01-16 Last updated: 2024-09-02Bibliographically approved
2. Multidimensional Panel Data Regression Model: The Case of the Multidimensional Home Ownership Vacancy Rate in the USA
Open this publication in new window or tab >>Multidimensional Panel Data Regression Model: The Case of the Multidimensional Home Ownership Vacancy Rate in the USA
(English)Manuscript (preprint) (Other academic)
Abstract [en]

We expanded the use of structured machine learning in regression for nowcasting by utilizing multidimensional panel data. Our primary goal was to predict home ownership vacancy rates across various states and metropolitan statistical areas, especially when key economic data are sampled at mixed frequencies. We successfully employed our proposed extended multidimensional machine learning panel data model to forecast the three-dimensional home ownership vacancy rate in the United States. The results suggest that our extended multidimensional time series regression model is very useful for nowcasting/forecasting home ownership vacancy rates and for performing better than does the traditional time series regression model. Our results are general, and our extended multidimensional time series regression model can be applied to any multidimensional macroeconomic problem.

Keywords
Multidimensional panel data, MIDAS, home ownership vacancy rate, regularized regression sg-LASSO
National Category
Probability Theory and Statistics
Identifiers
urn:nbn:se:hj:diva-65982 (URN)
Funder
The Jan Wallander and Tom Hedelius Foundation, H2021-0027
Note

Included in doctoral thesis in manuscript form.

Available from: 2024-08-19 Created: 2024-08-19 Last updated: 2024-09-02
3. MIDAS Poisson regression advancements in dengue fever prediction using Google Trends and environmental data
Open this publication in new window or tab >>MIDAS Poisson regression advancements in dengue fever prediction using Google Trends and environmental data
(English)Manuscript (preprint) (Other academic)
Abstract [en]

This paper proposes a new approach for the generalization of Poisson regression for count time series data typically sampled at different frequencies to improve the forecasting accuracy of dengue cases in Pakistan. Mixed data sampling (MIDAS) was introduced in the context of Poisson regression because it enables us to model and forecast dengue events more accurately. Different polynomial weights selected from the literature are applied in the MIDAS and U-MIDAS settings, and different forecast combinations are used to improve the forecasting accuracy of dengue event counts. For the 2006–2017 period, the proposed model correctly forecasts significantly more dengue cases than does the standard Poisson regression model for all forecasting horizons. Furthermore, Google Trends data can be a usefuladdition to traditional numeric data for forecasting dengue cases.

Keywords
Dengue, Climate, Google Trends, U-MIDAS, MIDAS Poisson, Count Data, MIDAS-PRM
National Category
Probability Theory and Statistics
Identifiers
urn:nbn:se:hj:diva-65983 (URN)
Note

Included in doctoral thesis in manuscript form.

Available from: 2024-08-19 Created: 2024-08-19 Last updated: 2024-09-02
4. Forecasting Financial Volatility: A Dual-Parameter Heterogeneous Autoregressive Model for Realized Volatility
Open this publication in new window or tab >>Forecasting Financial Volatility: A Dual-Parameter Heterogeneous Autoregressive Model for Realized Volatility
(English)Manuscript (preprint) (Other academic)
Abstract [en]

The heterogeneous autoregressive (HAR) model, applied to the realized volatility (RV) of financial time series, is specifically designed to capture both the persistent memory and the time-varying dynamically evolving nature of volatility clustering characteristics in financial time-series data. Typically, HAR models establish a linear relationship linking the current and past RVs using ordinary least squares (OLS) for estimation. However, this methodology might exhibit reduced reliability when RV data encounter abrupt and enduring shocks. To address this issue, we introduce a two-parameter shrinkage estimator for the HAR-RV model, incorporating two additional variables for sudden shocks to capture the broader spectrum of market behavior and conditions. This estimator effectively mitigates overparameterization issues and exhibits superior properties in scenarios involving sudden market changes and underlying trends. This enhancement not only improves the model’s goodness of fit but also improves the accuracy of out-of-sample forecasts. We apply our estimator to forecast financial time series, with the empirical results demonstrating superior forecasting performance in terms of improved out-of-sample R² values, corroborating our theoretical insights andsimulation studies.

Keywords
Shrinkage estimation, heterogeneous autoregressive-realized volatility, HAR-RV, S&P 500, forecasting
National Category
Probability Theory and Statistics
Identifiers
urn:nbn:se:hj:diva-65984 (URN)
Note

Included in doctoral thesis in manuscript form.

Available from: 2024-08-19 Created: 2024-08-19 Last updated: 2024-09-02

Open Access in DiVA

Kappa(3103 kB)36 downloads
File information
File name FULLTEXT01.pdfFile size 3103 kBChecksum SHA-512
c60a6c379caa482166881b6936c2ff059e579c3ef23883528723a97cfd300c4208c19986a322aa292857b3d9d3304c704de3d77ae061253edd25a9f45483804c
Type fulltextMimetype application/pdf

Authority records

Omer, Talha

Search in DiVA

By author/editor
Omer, Talha
By organisation
JIBS, Statistics
Probability Theory and Statistics

Search outside of DiVA

GoogleGoogle Scholar
Total: 36 downloads
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

isbn
urn-nbn

Altmetric score

isbn
urn-nbn
Total: 731 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf