Volatility forecasting using Deep Learning and sentiment analysis V Ncume1 T. L van Zyl2000000034281630X and A

2025-05-06 0 0 490.75KB 13 页 10玖币

侵权投诉

Volatility forecasting using Deep Learning and

sentiment analysis

V Ncume1, T. L van Zyl2[0000−0003−4281−630X], and A

Paskaramoorthy3[0000−0002−7812−5909]

1Computer Science and Applied Mathematics, University of the Witwatersrand,

Johannesburg, South Africa

vuyoncume68@gmail.com

2Institute for Intelligent Systems, University of Johannesburg, Johannesburg, South

Africa

tvanzyl@gmail.com

3Department of Statistical Sciences, University of Cape Town, Cape Town, South

Africa

ab.paskaramoorthy@gmail.com

Abstract. Several studies have shown that deep learning models can

provide more accurate volatility forecasts than the traditional methods

used within this domain. This paper presents a composite model that

merges a deep learning approach with sentiment analysis for predicting

market volatility. To classify public sentiment, we use a Convolutional

Neural Network, which obtained data from Reddit global news headlines.

We then describe a composite forecasting model, a Long-Short-Term-

Memory Neural Network method, to use historical sentiment and the

previous day’s volatility to make forecasts. We employed this method on

the past volatility of the S&P500 and the major BRICS indices to cor-

roborate its eﬀectiveness. Our results demonstrate that including senti-

ment can improve Deep Learning volatility forecasting models. However,

in contrast to return forecasting, the performance beneﬁts of including

sentiment for volatility forecasting appears to be market speciﬁc.

Keywords: Deep Learning ·Support Vector Regression ·Generalized

Autoregressive Conditional Heteroskedasticity ·Volatility Forecasting

1 Introduction

Deep Learning has shown to be useful in sequential data prediction tasks such as

time series forecasting and text prediction. Given suﬃcient compute power and

time, Deep Learning algorithms are able to learn from large datasets and out-

perform traditional machine learning and statistical techniques. Consequently,

there is increasing interest in using Deep Learning for economic and ﬁnancial

forecasting owing to its successes in other domains. A growing literature in-

vestigates whether Deep Learning algorithms with various architectures can be

used to make predictions in ﬁnancial markets that can be exploited for proﬁt

[12, 13, 16, 17].

arXiv:2210.12464v2 [cs.LG] 17 Nov 2022

2 V Ncume et al.

Previous work in the ﬁnancial time series forecasting domain has acknowl-

edged the importance of sentiment in predicting ﬁnancial markets, and thus we

seek to use sentiment data in conjunction with a deep learning model to increase

prediction accuracy. Text from the internet is increasingly becoming more rel-

evant as an important type of data to be included in predictive models. For

example, [11] develops a prediction model that combines news events and ﬁnan-

cial data to predict the ﬂuctuation of foreign currency. [5] shows that opinions on

popular online platforms are strong predictors of earnings surprises and future

market returns for stocks. Other studies have corroborated that social media

posts are useful for prediction in ﬁnance (for example, [25] and [23]).

Deep learning techniques have been used to forecast market returns in various

ways and have shown to be more accurate at making predictions when sentiment

is included as an input. For example, In [18] and [20], a Long Short Term Memory

Neural Network (LSTM) is used to forecast the stock closing price along with

data from Twitter to gauge public sentiment. [12] proposes a hybrid algorithm

where a CNN is used for classifying sentiments, which were used as inputs into

an LSTM Neural Network to predict stock prices with similar results to [18] and

[20].

Whilst forecasting market returns is receiving increased attention, using Deep

Learning models for volatility forecasting (another important problem in ﬁnance)

has been largely unexplored. Volatility forecasting can be seen as easier than re-

turn forecasting due to the presence of second-order autocorrelation in empirical

returns (known as “volatility clustering“). Volatility is typically modeled using

traditional time series models, such as the Generalized Autoregressive Condi-

tional Heteroskedasticity (GARCH) model or its extensions [2]. However, studies

by [15], [9], [21] and [24] show that Deep Learning methods can outperform the

more traditional methods in the volatility forecasting domain.

The study by [9] shows that there is still a large gap between the state-of-the-

art deep learning techniques available and their use in the volatility forecasting

domain. We look to close this gap by proposing a hybrid deep learning model

which forecasts sentiment, which is used in turn to forecast the volatility of a

market index. More speciﬁcally, we combined a Convolutional Neural Network

(CNN) for sentiment analysis and a Long Term Short Term Memory [8] Neural

Network for the volatility predictions. We used this hybrid approach to forecast

the past volatility of the S&P500 and the major BRICS indices [4]. Our approach

is similar to [12], except that they apply their method to return forecasting,

whilst we are concerned with volatility forecasting.

2 Background And Related Work

The volatility forecasting problem is, at its core, a regression problem and there

are various methods that can be used to model the data and make forecasts.

The objective of a forecasting model for volatility prediction using nonlinear

regression techniques is to form a relationship of the following form:

y=f(xn)(1)

Volatility forecasting using Deep Learning and sentiment analysis 3

where xn= (x1, . . . , xn)is an input vector and yis the output value (the volatil-

ity). In our problem, the previous volatility and returns are used as inputs.

The function fis found by using training data to select the regression model’s

parameters to minimize empirical loss between the model outputs and the actual

outputs. Commonly, empirical loss is deﬁned by the sum of squared errors. For, in

an ordinary least squares regression problem with a single predictor, the function

f=w0xis linear, and the ﬁtting problem is deﬁned as:

minw

i=n

(yi−wixi)2(2)

where iis an index variable for the data in the training sample, which has size

2.1 SVR

The objective in Support Vector Regression (SVR) is to minimize the size of the

coeﬃcient vector (measured by its `2-norm), whilst requiring that the predictive

accuracy of the model (measured by its `1-norm) is at most . Compared to OLS,

the prediction error of the model is thus treated like a constraint. We can tune

the maximum error allowable to obtain accuracy desired. The constraints and

objective function thus become:

Minimize: 1

2||w||2(3)

such that:

|yi−wixi| ≤ (4)

where i= 1, . . . , n represents the index for each datapoint in training data.

It is possible that for various pairs (xi, yi)there is no solution withat ensures

the prediction error is less than . Thus, to ensure the feasibility of the optimiza-

tion problem, slack variables ξi, ξ∗

ican be included in the problem speciﬁcation

in the following manner [22]:

Minimize:

2||w||2+C

i=1

(ξi+ξ∗

i)(5)

such that:

yi−wixi≤+ξi

wixi−yi≤+ξ∗

ξi, ξ∗

i≥0.

Here, Cis a hyper-parameter that controls the trade-oﬀ between the size of the

coeﬃcient vector and the tolerance of errors larger than .

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

VolatilityforecastingusingDeepLearningandsentimentanalysisVNcume1,T.LvanZyl2[000000034281630X],andAPaskaramoorthy3[0000000278125909]1ComputerScienceandAppliedMathematics,UniversityoftheWitwatersrand,Johannesburg,SouthAfricavuyoncume68@gmail.com2InstituteforIntelligentSystems,UniversityofJohannesburg...

展开>> 收起<<

Volatility forecasting using Deep Learning and sentiment analysis V Ncume1 T. L van Zyl2000000034281630X and A.pdf

共13页,预览3页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

Volatility forecasting using Deep Learning and sentiment analysis V Ncume1 T. L van Zyl2000000034281630X and A

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: