Modeling Order Book Fluctuation by Monte Carlo Technique

CONTENTS

Page no. 1) Certificate 2 2) Acknowledgement 3 3) Abstract 5 4) Introduction 6 5) Simulation code 8 ➢ Order Book 8 ➢ Diffusion 9 ➢ Price and Annihilation 11 ➢ One Trade return 14 ➢ Waiting time between consecutive trades 16 ➢ Conditional return 19 ➢ Hurst curve 20

6) Results and Discussion 22 7) Summary 28 8) Future Prospects 29 9) References 30

ABSTRACT

This Project aims to simulate Order Book Fluctuations in a typical Stock Exchange using random variables. We aimed to simulate Bak, Paczuski and Shubik model using ANSI C and compiler Code-Blocks. We Simulated the Order Book using a two dimensional array to get better accessibility and file handling. We used the random function generator to generate random variables .

INTRODUCTION

Order Book : An order book is the list of orders (manually and now electronically) that a trading venue (in particular stock exchanges) uses to record the interest of buyers and sellers in a particular financial instrument.

Trading Engine : A trading engine uses the book to determine which orders can be fulfilled i.e. what trades can be made. Trades are done on the basis of matching between bids and asks.

Order Book Simulation : Order Book Simulation is used to predict the market behavior and estimate price based on various parameters like random variables used etc.

Need of Simulation : If we say that we simulate order book using random variables , then what is the predictability and usage of results in different environment or with different random variables. The question was first addressed by Hurst by inventing a new statistical method called Rescaled Range Analysis in which randomly generated data was analyzed and he got to the conclusion that even randomly generated data can have long range memory effect in some processes and data can be estimated on the basis of same technique. We simulate the order book to see that how predictable is the outcome on the basis of input given and see auto correlation between the two sets.

These simulation techniques steer modern investment portfolios of various investment firms , investment banks and Hedge funds. Now days simulation results are being provided by various consultancy firms to private investors for better investment strategies.

Bak , Paczuski and Shubik Model [31]

On a line representing the price axis, two kinds of particles are placed. The first kind, denoted A (ask), corresponds to sell orders, while the second, B (bid), corresponds to buy orders. The position of the particle is the price at which the order is to be satisfied. A trade can occur only when two particles of opposite type meet. If that happens, the orders are satisfied and the particles are removed from the system. This can be described as annihilation reaction A + B → ∅. It is evident that all B particles must lie on the left with respect to all A particles. The particles diffuse freely and in order to keep their concentration constant on average, new orders are inserted from the left (B type) and from the right (A type). The whole picture of this order-book model is therefore identical to the two-species diffusion-annihilation process. The changes in the price are mapped on the movement of the reaction front.

In this model there are N particles of each type, i.e. total 2N particles placed at the segment of length L. The particles can occupy integer positions from the set {1, 2, . . ., L}. In one update step we choose one particle and change its position as ci’=ci+1 or ci’=ci -1 (there is no bias, so both signs of the change have the same probability), on condition that the new position stays within the allowed interval, 1 ≤ ci’ ≤ L. We use the convention that the time advances by 1/(2N) in one step. If the new site was empty or there was already another particle of the same type at the new position, nothing more happens and the update is completed. We set ci(t + 1/(2N)) = ci’ and ck(t + 1/(2N)) = ck(t), k!=i . On the other hand, if the new site is occupied by a particle of opposite type, say, particle j, so that cj(t) = ci’, then the two particles annihilate. To keep the number of particles constant, we immediately supply two new particles at opposite edges of the allowed segment. E.g. if i was type B and j was type A, the update is ci(t + 1/(2N)) = 1, cj(t + 1/(2N)) = L and ck(t + 1/(2N)) = ck(t), k!= i, j. The annihilation corresponds to an elementary transaction.
The price set in this deal is just the position where the annihilation took place, x(t + 1/(2N)) = ci’. If the transaction did not occur, the price stays unchanged, x(t+1/(2N)) = x(t). This completes the definition of the variant of the BPS model simulated here.

SIMULATION CODE

Programing Language Used : ANSI C

Compiler Environment : Code Blocks

Total Number of particles in the system : 200 A type and 200 B type

Order book : The order is simulated as a two dimensional array of size[2N][L] as this simplifies the problem statement to a large extent. The column number represents the value of the element placed in it.

A sample array is shown below for reference :

1 |5 |5 |5 |5 |5 |5 |5 | |5 |1 |5 |5 |5 |5 |5 |5 | |5 |5 |1 |5 |5 |5 |5 |5 | |5 |5 |5 |1 |5 |5 |5 |5 | |5 |5 |5 |5 |9 |5 |5 |5 | |5 |5 |5 |5 |5 |9 |5 |5 | |5 |5 |5 |5 |5 |5 |9 |5 | |5 |5 |5 |5 |5 |5 |5 |9 | |

5 = Background element , 1 = B Type , 9 = A Type

All the background elements are taken as 5 to avoid ambiguities using conditional statements in C . All B type elements are taken as 1’s and all A type elements are taken as 9’s. Since all the elements occupy a unique row, horizontal diffusion will not change the number of particles in each row.

Initiating Array : The first step in the program is to initiate the 2N*L array and put all the elements at places as described above. Function written to initiate the array is given below

void set1(){

if(n