Periodic Reporting for period 1 - NONCAUSALBubble (Noncausal time series models for the forecasting of speculative bubbles)
Período documentado: 2020-07-01 hasta 2022-06-30
Exuberant increases in the price of certain goods or financial assets, far beyond what could appear to be a reasonable intrinsic value, are commonly designated as “bubbles”. Such phenomena are not new: the Tulip Mania of the 17th century and the South Sea bubble of 18th century, for instance, are well documented historical instances of such events. Sharp reversals, or “crashes”, systematically follow the exuberant increasing phase of bubbles and have dramatic impact on the broader economy and society. The United States housing bubble, which peaked in 2006, was the trigger for a global economic crisis which required unprecedented intervention by states and central banks across the world.
Monitoring and forecasting the evolution of asset prices, especially the ones undergoing rapid increases, could provide the ability for financial regulators to steer the markets away from overheating, avoiding sharp collapses or allowing to contain their impact. Forecasting bubbles and their crashes remains however an open research question.
A recent modelling approach at the intersection of time series econometrics, statistics and probability theory -so-called “anticipative” or “noncausal” time series models- is very promising in that regard as it allows to fit and reproduce adequately the observable and statistical characteristic of bubbles across a wide range of financial indexes, stocks, commodities, cryptocurrencies and economic indicators. A blind spot of this non-standard modelling approach is that no theory exists regarding forecasting the future evolution of bubbles or the incoming occurrence of a crash. The objective of this project is precisely to provide such theoretical results which will enable the use of anticipative time series models for forecasting bubble crashes.
Monitoring and forecasting the evolution of asset prices, especially the ones undergoing rapid increases, could provide the ability for financial regulators to steer the markets away from overheating, avoiding sharp collapses or allowing to contain their impact. Forecasting bubbles and their crashes remains however an open research question.
A recent modelling approach at the intersection of time series econometrics, statistics and probability theory -so-called “anticipative” or “noncausal” time series models- is very promising in that regard as it allows to fit and reproduce adequately the observable and statistical characteristic of bubbles across a wide range of financial indexes, stocks, commodities, cryptocurrencies and economic indicators. A blind spot of this non-standard modelling approach is that no theory exists regarding forecasting the future evolution of bubbles or the incoming occurrence of a crash. The objective of this project is precisely to provide such theoretical results which will enable the use of anticipative time series models for forecasting bubble crashes.
A single-authored preprint (94 pages, mathematical proofs included) with new theoretical results supported by computer simulation experiments and illustrated with financial data application was submitted to the top field journal “Journal of Business and Economic Statistics” (top 98% according to Article Influence Percentile metric) and underwent 3 rounds of peer-review. The preprint was finally accepted for publication by the journal in July 2021. The publication is uploaded in the repository: https://zenodo.org/record/6519100(se abrirá en una nueva ventana).
The results were disseminated at several workshops, invited seminars and conferences, i.a.:
- 19th Econometrics Day of Nanterre University, 18th of November 2020, dedicated to Recent Developments in Applied Financial Econometrics
- Econometrics and Data Science department’s seminar, Vrije Universiteit Amsterdam, December 3rd 2020
- CREST Financial Econometrics Seminar, December 10th 2020
- Osaka University GSE-OSIPP Joint Seminar in Economics, December 17th 2020
- Conference on Financial Econometrics, December 19th 2020
- RIKEN High-dimensional Statistical Modeling Team Seminar, May 11th 2020
The results were disseminated at several workshops, invited seminars and conferences, i.a.:
- 19th Econometrics Day of Nanterre University, 18th of November 2020, dedicated to Recent Developments in Applied Financial Econometrics
- Econometrics and Data Science department’s seminar, Vrije Universiteit Amsterdam, December 3rd 2020
- CREST Financial Econometrics Seminar, December 10th 2020
- Osaka University GSE-OSIPP Joint Seminar in Economics, December 17th 2020
- Conference on Financial Econometrics, December 19th 2020
- RIKEN High-dimensional Statistical Modeling Team Seminar, May 11th 2020
The peer-reviewed publication is the first to provide theory for the forecasting of the so-called anticipative time series models. Among other results, the odds of future crash of ongoing exuberant price increase in the framework of these models are now quantifiable. Using existing standard statistical procedures to estimate key parameters from the data, it is now possible to apply these models to provide crash odds forecasts of bubbles, as illustrated by an example on financial data in one of the preprint’s version. This opens an avenue for central banks, financial regulators, risk managers and financial practitioners. New ongoing research projects have been launched on the basis of applying the novel theory.
Beyond the forecasting ability provided, the results also shed new light on the economic mechanisms put forth by the literature to explain how bubble arise. A fruitful line of research dating back to the 1980s argues that bubbles arise on financial markets because it may be rational for speculators and investors to participate in its inflation – rational in the sense that the potential gains from betting on the bubble to continue increasing in the near-future outweighs the risk of a collapse. In the 2000s however came empirical evidence against the classical rational bubble models. Intuitively, the classical models predicted that bubbles shall be prone to climb towards extremely high levels, with such degree of extremeness not being warranted by observed past bubbles. New results in the project’s publication show that anticipative models extend classical rational bubble models while at the same time being compatible with the empirical evidence.
Beyond the forecasting ability provided, the results also shed new light on the economic mechanisms put forth by the literature to explain how bubble arise. A fruitful line of research dating back to the 1980s argues that bubbles arise on financial markets because it may be rational for speculators and investors to participate in its inflation – rational in the sense that the potential gains from betting on the bubble to continue increasing in the near-future outweighs the risk of a collapse. In the 2000s however came empirical evidence against the classical rational bubble models. Intuitively, the classical models predicted that bubbles shall be prone to climb towards extremely high levels, with such degree of extremeness not being warranted by observed past bubbles. New results in the project’s publication show that anticipative models extend classical rational bubble models while at the same time being compatible with the empirical evidence.