Research Repository

Natural Time and Crash Risk

Turkoglu, Ata (2015) Natural Time and Crash Risk. PhD thesis, University of Essex.

Natural Time and Crash Risk.pdf

Download (4MB) | Preview


The deviation of financial returns from normal distribution is a well‐documented stylized fact. Nonetheless, finance professionals and investors alike pay attention to these deviations almost only when a crisis erases years’ worth of gains. And despite decades’ worth of literature, the culprit for non‐normal distribution of financial returns is still not determined with certainty. In this research, I address the non‐normality of return distributions and financial crashes together. Specifically, I aim to identify the determinants of non‐normality in a high frequency setting and utilize these variables to forecast financial crashes. To this effect, multiple instruments and time horizons are considered. The contribution of this thesis is multifold. The “natural time” approach introduced here, uses order book variables to achieve normally distributed high frequency returns via subordination. In its essence, natural time is a two‐step procedure which uses high frequency order book variables as a gauge for variance while sampling in transaction time. Natural time provides the reader with a new lens to view the financial markets and underscores two important aspects of the high frequency world; sampling frequency affects the distributions we observe and order book variables such as liquidity are the key to heteroscedasticity in asset returns. So much so that subordination with order book variables under transaction time achieves the normal return distribution which underlies numerous financial theories we use today. I further extend the use of these order book variables by introducing the “market heat” metric. Market heat generates successful binary flash crash predictions and its success adds support to the claim that liquidity concerns may be the primary driver of price formation processes. Finally, I expand the findings of this research on high frequency asset returns to a macroeconomic setting by producing currency devaluation predictions for G10 currencies. The early warning systems produced here demonstrate that not only debt related macroeconomic variables but also liquidity related market variables are at play when it comes to currency fluctuations.

Item Type: Thesis (PhD)
Divisions: Faculty of Science and Health > Computer Science and Electronic Engineering, School of > Centre for Computational Finance and Economic Agents
Depositing User: Ata Turkoglu
Date Deposited: 26 Feb 2016 13:21
Last Modified: 23 Feb 2021 02:00

Actions (login required)

View Item View Item