Intelligent Weighted Fuzzy Time Series Model For

Financial Markets Forecasting

Ruixin Yang, Paul Jones and Nagiza F. SamatovaLAS & Big Data Analytics Group, Computer Science Department, NC State University

Contact: Ruixin Yang(ryang9@ncsu.edu) and Nagiza Samatova (samatova@csc.ncsu.edu)

Motivation

● Risk management ── to minimize, financial security monitor, and control the probability and/or impact of unfortunate events.

How to be Intelligent?

Forecast Performance

Why Fuzzy Time Series Model? [1,2]

Financial security is critical for both national and individual level. A sound and stable financial system is prerequisite for sustainable economic growth and thus risk forecast is playing a crucial role in modern financial analysis. Even a tiny improvement in markets forecasting accuracy may have a huge impact on decision making for:

● Portfolio investment ── investments in the form of a group of assets, including transactions in equity securities and debt securities.

The advantages of Fuzzy Time Series (FTS) Models:● Naturally universal approximated (can represent any

complex financial models if well designed).● Linguistic expressions to describe daily observations so that

human experience and knowledge can be integrated. ● Ability to deal with vague data with uncertainty and

nonlinearity

Challenge:● Several parameters such as interval length, weight factors impact the forecast performance● How to set these parameters intelligent other than markets by markets?Approach:● Using a novel evolutionary algorithm ── Human Learning Optimization (HLO)● Each parameter can be globally optimized by HLO

Individual Learning

Random learning

Social learning

Fuzzy logic relationships

Long-term tendency

Interval length andweight factors

Our model shows state-of-the-artaccuracy on four world leading markets:● Dow Jones Index(DJI)

● German Stock Index (DAX)

● Japanese Stock Index (NIKKEI)

● Taiwan Stock Index (TAIEX)

Weighted Fuzzy Time Series Model● Each data point has a corresponding fuzzified number (such as Ai, Aj)

which is determined by the interval length and universe discourse.

● Fuzzy logic relationships (FLRs) are generated by time sequence as:Ai ──> Aj

Ai+m ──> Aj+m+1

Ai+n ── > Aj+n+1

Ai+m

Aj

Ai

Aj+m+1

Aj+n+1

Ai+n

● Since the fuzzified number may have several logic relationships, fuzzy logic relationship groups (FLRGs) are generated as:

Ai ──> A1, A2, ..., AkChallenge:● Does each data point has same effect for current forecast?● How to handle the situation where index reaches record high or low?Approach:● Add weight to each data point using chronological-order and volume.● Incorporate a long-term tendency (LT) with FLRs by the jump theory.

e.g. if Ai ──> Aj, the jump will be j-i and LT is calculated by all jumps.

Linear combination of two strategies

Final forecast

Our efforts attempt to design a new framework to improve forecast accuracy for better financial decision making.

[1] “An Intelligent Weighted Fuzzy Time Series Model Based on a Sine-Cosine Adaptive Human Learning Optimization Algorithm and Its

Application to Financial Markets Forecasting”, Yang/Xu/He/Ranshous/Samatova, ADMA 2017.

[2] “An Intelligent and Hybrid Weighted Fuzzy Time Series Model Based on Empirical Mode Decomposition for Financial Markets

Forecasting”, Yang/He/Xu/Ni/Jones/Samatova, Submitted to SDM 2018.

Causal

Models

Quantitative

Forecasting

SVM

Models

Time Series

Models

Moving

Average

Exponential

Smoothing

Trend

ModelsFuzzy Time

Series Models

Easily get trapped

in outliers or

influential cases.

Usually over-fitting

from optimizing the

parameters to model

selection.

Vulnerable to false signals

and getting whipsawed

back and forth.

Cannot handle trends well

so lags any trend in the

data and forth.

Any indicator is unable to

detect whether the

profitable move of the

market is a short lived one

or a major.