Modeller for værvarsler om finansiell usikkerhet

Xingyi Li disputerer for ph.d.-graden med avhandlingen "Essays on the Horizon of Volatility Predictability and Volatility Model Validation" tirsdag 7.januar 2020.

Hun har fulgt doktorgradsprogrammet ved Handelshøyskolen ved UiA.

Slik oppsummerer Xingyi Li avhandlingen med egne ord:

Financial Volatility Predictability and Model Validation

Volatility is a measure of uncertainty about financial returns. Modelling and forecasting volatility has become a rapidly growing interest and crucial task in financial industry since the early sixties. For forecasting, this dissertation answers how far into the future volatility forecasts can be made (the first and third essays) and how volatility predictability varies across the market states (the second essay). For modelling, this dissertation examines how well volatility models capture stylized facts or dynamics observed in financial markets (the fourth essay).

The first essay uses the model-based approach to explore the horizon of volatility predictability. It quantifies the forecast accuracy across horizons by introducing the notions of the spot and forward predicted volatility. These notions contribute to describe the term structure of volatility predictability and provide more deep information on the horizon of volatility predictability and forecast accuracy than previous studies. The results of this essay suggest that the horizon of volatility predictability is substantially larger than that reported in previous studies.

The stock market states are classified into bull and bear markets in the second essay. This essay investigates the horizon of volatility predictability and quantifies the volatility forecast accuracy across the bull and bear market states separately. Both the first and second essays use the same forecast accuracy measure. The general conclusion of the essay is that the volatility predictability is strongest in bad economic times represented by bear market states. That is, volatility predictability is best when it is most needed, during periods of high turbulence and uncertainty.

The third essay investigates the same question as the first essay using model-free tests. The name “model-free” comes from the fact that the conclusion on the horizon of volatility predictability is drawn directly from the observed returns without any volatility models involved. The essay significantly extends the horizon of volatility forecastability reported in the earlier model-free studies.

A common research novelty in the first three essays is that they, for the first time, adopt the so-called “meta-analysis” to aggregate the results on volatility predictability from individual assets. A key benefit of this approach is that the aggregation of information on individual assets leads to a higher statistical power and more robust forecast accuracy estimates than it is possible to obtain from the information on any individual asset.

The fourth essay introduces a novel framework which answers the question of how well a volatility model reproduces the stylized facts. Specifically, the framework encompasses a list of statistics quantifying the volatility stylized facts. The empirical study of the essay demonstrates a model validation procedure that tests the ability of a volatility model to replicate the list of statistics.  The main conclusion in this essay is that none of the tested models is able to fully replicate the volatility stylized facts.