Grant: Discovery Grants Program - Individual
York University
$18,966.00
Advances in Inference Methods for Stationary Martingales and Non-Gaussian Processes
FY 2024-25
Summary
This research proposal introduces new inference methods for nonlinear dynamic processes with independent, identically, and non-normally distributed errors. Among these processes are stationary martingales and causal-noncausal (mixed) processes with local explosive patterns, such as short-lasting (local) trends, spikes, and bubbles characterizing, for example, the time series of oil and energy prices, commodity prices, and digital currency rates. Applied researchers in Economics and Finance, are interested in modelling and forecasting these time series and their co-movements with macroeconomic variables. This research will provide them with new inference methods for testing hypotheses on the specification of dynamic nonlinear models for these processes. The proposed inference methods will be based on nonlinear autocovariance conditions and the Generalized Covariance (GCov) estimator. The GCov is a one-step semi-parametric estimator, which does not require prior knowledge of the error distribution, often being difficult to determine in practice. In addition, it has desirable asymptotic properties, and its objective function leads to a portmanteau test statistic with a known limiting distribution allowing for hypothesis testing. The finite sample performance of the new tests will be assessed to verify their reliability and adjust their robustness in various applications, including high dimensional data. In practice, identification problems may arise when, for example, the error distribution is close to the normal distribution. To address these issues, a new set identification approach to inference under partial identification will be developed. Special attention will be given to the mixed vector autoregressive (VAR) processes providing a convenient alternative to the linearized macroeconomic models, like the Canadian LENS (Large Empirical Semi-structural) model of the Bank of Canada for policy making. The mixed VAR processes can accommodate the local explosive patterns of a component series and do not require stringent identifying assumptions. An estimator of the forecasting density of the mixed VAR model will be introduced to allow practitioners to forecast from these models and assess the forecast accuracy. The difficulty in forecasting the future values on a bubble, spike, or local trend will be addressed by introducing inference methods leading to robust prediction intervals and reliable coverage rates. Since the processes of interest are time-irreversible, a back-casting method will also be developed. The proposed research will advance the statistical inference methodology and offer ample opportunity to train graduate students in time series analysis. It will be conducted with a strong commitment to the EDI guidelines, and efforts will be made to involve a diverse group of students in this research. The outcomes will be disseminated to academics and practitioners and published in high-quality outlets promoting fair and diverse representation.
Program
Discovery Grants Program - Individual
Recipient
Location
Ontario, Canada
Recipient Type
Academia
Type
Grant
Source
NSERC's Awards Data
As published
Application id
RGPIN-2024-06211
Name nom
Jasiak, Joann
Department département
Economics
Organization id
38
Institution établissement
York University
Province en
Ontario
Province fr
Ontario
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2024
Competition year année de concours
2024
Award amount
18966
Program id
RGPIN
Program name en
Discovery Grants Program - Individual
Program name fr
Programme de subventions à la découverte - individuelles
Group en
DISCOVERY RESEARCH
Group fr
RECHERCHE AXÉE SUR LA DÉCOUVERTE
Committee code
1508
Committee name en
Mathematics and Statistics
Committee name fr
Mathématiques et statistique
Area of application code
1212
Area of application group en
Advancement of knowledge
Area of application group fr
Avancement des connaissances
Area of application en
Social sciences
Area of application fr
Sciences sociales
Research subject code
8000
Research subject group en
Social sciences and humanities
Research subject group fr
Sciences humaines
Research subject en
Social sciences and humanities
Research subject fr
Sciences humaines
Field of research list names en
Not available
Field of research list names fr
Non disponible
Socioeconomic objective list names en
Not available
Socioeconomic objective list names fr
Non disponible
Application title
Advances in Inference Methods for Stationary Martingales and Non-Gaussian Processes
Keywords
time series analysis; semi-parametric inference; stationary martingale; noncausal process; non-gaussian process; time-irreversible process; independence test; nonlinear model; portmanteau statistic; nonlinear autocovariance
Application summary
This research proposal introduces new inference methods for nonlinear dynamic processes with independent, identically, and non-normally distributed errors. Among these processes are stationary martingales and causal-noncausal (mixed) processes with local explosive patterns, such as short-lasting (local) trends, spikes, and bubbles characterizing, for example, the time series of oil and energy prices, commodity prices, and digital currency rates. Applied researchers in Economics and Finance, are interested in modelling and forecasting these time series and their co-movements with macroeconomic variables. This research will provide them with new inference methods for testing hypotheses on the specification of dynamic nonlinear models for these processes. The proposed inference methods will be based on nonlinear autocovariance conditions and the Generalized Covariance (GCov) estimator. The GCov is a one-step semi-parametric estimator, which does not require prior knowledge of the error distribution, often being difficult to determine in practice. In addition, it has desirable asymptotic properties, and its objective function leads to a portmanteau test statistic with a known limiting distribution allowing for hypothesis testing. The finite sample performance of the new tests will be assessed to verify their reliability and adjust their robustness in various applications, including high dimensional data. In practice, identification problems may arise when, for example, the error distribution is close to the normal distribution. To address these issues, a new set identification approach to inference under partial identification will be developed. Special attention will be given to the mixed vector autoregressive (VAR) processes providing a convenient alternative to the linearized macroeconomic models, like the Canadian LENS (Large Empirical Semi-structural) model of the Bank of Canada for policy making. The mixed VAR processes can accommodate the local explosive patterns of a component series and do not require stringent identifying assumptions. An estimator of the forecasting density of the mixed VAR model will be introduced to allow practitioners to forecast from these models and assess the forecast accuracy. The difficulty in forecasting the future values on a bubble, spike, or local trend will be addressed by introducing inference methods leading to robust prediction intervals and reliable coverage rates. Since the processes of interest are time-irreversible, a back-casting method will also be developed. The proposed research will advance the statistical inference methodology and offer ample opportunity to train graduate students in time series analysis. It will be conducted with a strong commitment to the EDI guidelines, and efforts will be made to involve a diverse group of students in this research. The outcomes will be disseminated to academics and practitioners and published in high-quality outlets promoting fair and diverse representation.