Project description
Non-causal time series models for speculative bubble prediction
Speculative bubbles in financial markets can result in dramatic damages to portfolio performances and threaten the stability of the financial system. Autoregressive and moving average processes known as non-causal time series models have proved their capacity to reproduce standardised facts from speculative bubbles such as locally explosive trajectories. On condition that their dynamics are better understood, they will allow the formulation of predictions on future bubble trajectories. However, understanding regarding the prediction of non-causal processes remains limited. The EU-funded NONCAUSALBubble project will focus on the lack of theoretical foundations for the forecasting of strong non-causal processes. The project will be based on recent developments of extreme value and alpha-stable distribution theories. Analytical assessment of crash probabilities will lead to an intuitive prediction model of bubble identification.
Objective
Speculative bubbles on financial markets, viewed as short-term explosive deviations of prices from a typical historical level and ending in an abrupt correction, have become common events across all major asset classes. They can have a dramatic impact on portfolio performances, financial institutions solvability and can compromise the stability of the financial system. Because of their ability to reproduce stylized facts from speculative bubbles such as locally explosive trajectories, noncausal time series models -autoregressive (AR) and moving average (MA) processes with roots located inside the unit circle- have been at the center of a recent fast-emerging literature in econometrics and finance. Provided their dynamics is better understood, they will enable to formulate forecasts of future bubble trajectories. If rapid progress is being achieved on estimation and fitting problematics, prediction theory of noncausal processes remains particularly scarce and limited to special elementary cases – mostly the univariate noncausal AR(1) with independent and identically distributed Cauchy errors.
The NONCAUSALBubble project aims at specifically addressing the lack of theoretical foundations for the forecasting of heavy-tailed noncausal processes. Building on recent tools from extreme value and alpha-stable distribution theories, NONCAUSALBubble will characterise the conditional distribution of future paths given the past observed trajectory during explosive episodes for 1) higher-order and 2) multivariate noncausal ARMA models. Closed-form formulations of the predictive distribution during bubble episodes will be derived alongside analytical quantification of the crash odds, and an intuitive prediction framework in terms of bubble pattern-recognition will be developed.
The project is hosted by VU Amsterdam, one of the top research groups in time series econometrics and forecasting.
Fields of science
Programme(s)
Funding Scheme
MSCA-IF - Marie Skłodowska-Curie Individual Fellowships (IF)Coordinator
1081 HV Amsterdam
Netherlands