You are on page 1of 1

Nonlinear Regression Models

Chapter
8
INTRODUCTION
Nonlinear regression consists of models that have a nonlinear formulation
and cannot be linearized via transformations. One of the most “infamous”
nonlinear regression models is the I-Star Market Impact Model introduced
by Kissell & Malamut (1999). This model is used extensively for electronic,
algorithmic, and high frequency trading. See Kissell et al. (2004), Kissell &
Malamut (2006), or Kissell (2013) for an overview of this model and its
applications. The estimation of this model is also provided in Chapter 10
(Estimating Market Impact Models).
The I-Star model has form:
 
b ¼ a0 $X1a1 $X2a2 $X3a3 þ b1 X4a4 þ ð1  b1 Þ þ ε
Y

where,
Y ¼ market impact cost of an order. This is the price movement in the
stock due to the buying and selling pressure of the order or trade. It is
comprised of a temporary and permanent component.
X1 ¼ order size as a percentage of average daily volume to trade
(expressed as a decimal)
X2 ¼ annualized volatility (expressed as a decimal)
X3 ¼ asset price (expressed in local currency)
X4 ¼ percentage of volume and used to denote the underlying trading
strategy (expressed as s decimal)
The parameters of the model are: a0, a1, a2, a3, a4 and b1.
The error term of the model ε has normal distribution with mean zero and
 
variance v2, that is, εwN 0; v2 .
It is important to highlight here that this model is purely nonlinear and it is
not possible to transform the model into a linear formulation using any of
the techniques of previous chapters.

Algorithmic Trading Methods, Second Edition. https://doi.org/10.1016/B978-0-12-815630-8.00008-9


Copyright © 2021 Elsevier Inc. All rights reserved. 197

You might also like