Page 1

Abstract—This paper is motivated by the aspect of uncertainty in

financial decision making, and how artificial intelligence and soft

computing, with its uncertainty reducing aspects can be used for

algorithmic trading applications that trade in high frequency.

This paper presents an optimized high frequency trading system that

has been combined with various moving averages to produce a hybrid

system that outperforms trading systems that rely solely on moving

INTRODUCTION

S computational power and data base capacities increase,

there will always be a parallel increase in the availability

of more data and information for trading systems to process.

High-frequency trading, despite being a new area, has proved

to be a ripe field for future implementation of trading systems

that can make use of this type of data and develop high

frequency trading strategies. The fundamental problem that

this thesis aspires to solve is to improve algorithmic trading

systems by taking a common sense based approach reflected

by fuzzy logic a decision making mechanism. Therefore, the

decisions taken by an algorithmic trading system should be

significant rather than just precise. When analysing the

problem from this perspective, it appears that fuzzy logic, as a

reasoning mechanism, is an obvious choice. However, fuzzy

logic on its own is never sufficient. Fuzzy logic can provide

very good results when used in designing trading systems, yet

using a hybrid fuzzy system would provide a much better

optimised trading system. Various artificial intelligence and

signal processing mechanisms will have to be incorporated in

the trading algorithms implemented to yield very good results.

The used application will The second application to be utilised

will be an Adaptive Neuro Fuzzy Inference System (ANFIS)

for high-frequency trading.

Dr. Abdalla Kablan is a lecturer and researcher at the Department of

Banking and Finance in the Faculty of Economics, Management, and

Accounting. University of Malta. Abdalla.Kablan@um.edu.mt

Professor Joseph Falzon is the Head of the Department of Banking and

Finance and the Dean of the Faculty Economics, Management, and

Accounting at the University of Malta. Joseph.falzon@um.edu.mt

This is an expert system that combines fuzzy reasoning with

the pattern recognition capability of neural networks. A new

event based volatilitymodel will counter unnecessary input in

the training phase and will therefore be proposed. The

Intraday Seasonality Observation Model is greatly enhanced

by taking factors such as the volatility and scaling laws of

financial time series into account. Excess data is thus removed

as this enhanced model makes it possible to observe specific

events and seasonalities in the data. The overall performance

of the Adaptive Neuro Fuzzy Inference System has been

greatly improved due to the more accurate input/data provided

by the new event based volatility model.

This paper extends on the concept of ANFIS that has been

introduced in [1,5] to optimise moving average strategies that

are very widely used in the industry. In this paper, ANFIS is

utilised to learn from the input of high frequency data and

moving averages applied on them to perform predictions on

the next market movement.

Due to the random nature of financial times series, the

prediction and forecast is extremely complicated, the

predictability/accurate forecast of most financial time series

such as stock prices or indices is a highly contentious issue as

the efficient market hypothesis declares that the current price

takes into account/uses/makes use of/contains all available

market information/data. [14]. Moreover, a filtering process

should be applied in order to separate the substantial noise

created by/caused by financial time series from the signal [3].

Furthermore, the practice of performing calculations in real-

world scenarios for the standard deviation in fixed time,

employed by traditional volatility models, has revealed major

drawbacks [11]. Measuring volatility should be done from an

event-based perspective by analyzing observations made after

the event. In addition, making the right decisions having

analyzed all of the inputs from other blocks is key to

producing an accurate system [6,7]. Artificial intelligence and

soft computing can provide a major solution to the above.

ANFIS combines two important aspects of artificial

intelligence. Neuro-fuzzy systems successfully combine the

human-like reasoning process of fuzzy logic with the ability of

neural networking to identify data patterns [16,1]. This