Category Financial Econometrics and Empirical Market Microstructure

Revisiting of Empirical Zero Intelligence Models

Vyacheslav Arbuzov

Abstract This paper describes a zero-intelligence approach implementation for the modeling of financial markets. We construct a mechanism of order flow and market engine simulation. We analyze stylized facts to estimate the quality of our models. The research is based on a 1 month order and execution history data of the Moscow Exchange (MOEX) for one stock (JSC “Aeroflot”).

Keywords Daniels model • Market microstructure • Mike-Farmer model • Order flow • Stylized facts • Tail exponent • Zero-intelligence models

JEL Classification G15, G17

1 Introduction

Agent-based models play an important role in understanding the mechanisms of financial markets driven by the advances in technologies that allow the creation and calibration of complex and very detailed mo...

Read More

Probability of Default Models

Here, and later in the paper, the default is understood as one of the following signals for its registration:

• A bank’s capital sufficiency level falls below 2 %.

• The value of a bank’s internal resources drops lower than the minimum estab­lished at the date of registration.

• A bank fails to reconcile the size of the charter capital and the amount of internal resources.

• A bank is unable to satisfy the creditors’ claims or make compulsory payments.

• A bank is subject to sanitation by the Deposit Insurance Agency or another bank.

We propose a forecast probability of default (PD) model, which is based on the relationship between banks’ default rates and public information...

Read More

Backtesting StressGrades

Below we show several are early warning backtesting case studies on ETF’s representing major asset classes. We calibrated DStress for each ETF based on the largest daily drawdown dates (e. g., —9.6 % for SPY on December 1, 2008). If StressGrades are predictive, we would expect an escalation in PStress and decline and DStress as we approach the drawdown date (e. g., December 1 for SPY). In other
words, we would expect volatility to be high and rising before the peak endogenous stress events.

image216Again, for simplicity we use the Normal Distribution to calculate PStress in the case studies below.

Given that StressGrades are driven by volatility, we expect StressGrades to fail in predicting Black Swans, but to help in detecting Dragon Kings.

Read More

Sample Selection Bias in Mortgage Market Credit Risk Modeling

Agatha Lozinskaia

Abstract The mortgage crisis that started in the U. S. in 2007 and lasted until 2009 was characterized by an unusually large number of defaults on the subprime mortgage market. As a result, it developed into a global economic recession and placed the stability of the world banking system in jeopardy. Therefore, the issues of credit risk modeling showed the shortcomings of the current credit risk practice. Truncation, or partial observability, and simultaneous equations bias causes sample selection bias. As a result, parameter estimates are biased and inconsistent. Firstly, we provide an overview of current approaches in the mortgage literature to control for the sample selection bias correction, such as the Heckman model and bivariate probit model with selection...

Read More

Model Calibration

In order to understand this model’s properties and its advantages, it is necessary to analyze how the model can reflect real data conditions. Thus, there is an issue of calibration of the model and applicability of the model for the description of a real market situation.

The percolation model allows us to simulate price change distribution in a one­time step as a hypothetical situation of interaction of agents for a certain time interval. The task of calibration is to select values of parameters and receive the

image076

Fig. 4 Algorithm of reverse engineering calibration

model’s empirical distribution, which is similar to the real-world market distribution in terms of some pre-selected measures Wiesinger et al. (2010).

This is the method of reverse engineering in the context of financ...

Read More

Numerical Simulation of the MRW Process

The bottleneck of numerical simulation of the MRW process (8) is simulation of logarithmically correlated noise! Дг [k]. Simulation of the discrete Gaussian noise process with given autocorrelation function (covariance matrix) is a well – known problem and is subjected to a trade-off: exact simulation processes usually requires a lot of computation resources, and fast algorithms typically provide only approximated solution. The most known exact simulation method is based on the Cholesky or LU-decomposition of the covariance matrix into lower – and upper-triangle matrices (Davis 1987). Though this method is very efficient for short time-series, having computational complexity of O(N2) it is not suitable for simulation of long (e. g. N > 103) realizations...

Read More