Have a personal or library account? Click to login
Affect Indicators for Stock Market Forecasting Cover

Affect Indicators for Stock Market Forecasting

Central European Economic Journal
Volume 12 (2025): Issue 59 (January 2025)
By: Joanna Michalak  
Open Access
|Dec 2025

Figures & Tables

Figure 1.

Time series for Twitter volume and volume ofSource: own preparation
Time series for Twitter volume and volume ofSource: own preparation

Figure 2.

Time series for Twitter volume and close price ofSource: own preparation
Time series for Twitter volume and close price ofSource: own preparation

Figure A1.

Impulse response function for Apple in the period 01/01/2016–30/06/2016 (VAR (2))Source: own work
Impulse response function for Apple in the period 01/01/2016–30/06/2016 (VAR (2))Source: own work

Figure A2.

Impulse response function for Amazon in the period 01/01/2016–30/12/2017Source: own work
Impulse response function for Amazon in the period 01/01/2016–30/12/2017Source: own work

MAE results for Apple and Amazon closing prices with LSTM approaches (test dataset)

Full periodFirst half of 2016First half of 2017
IndicatorsAAPLAMZNAAPLAMZNAAPLAMZN
Base model131839949179258627176138673274137855173356700252350198
Fear125483563180665696164580851246518980211031605302579474
Anger125998230188602498172433139249338907171247523273984994
negative VADER128457127183451424180822334279399172176222412229997950
negative NRC129982302181365827173876262256298833174086104208780619
Neutral VADER126443868186080492180506883256098881239225325256152057
Positive NRC132827363180629995176500689271267606193072633188347131
positive VADER133481260183951456190247008241919926149474564199168532
Sadness133854686181299519186525792259533810176089131209214699
Twitter volume12899904218614142720424486126393241216610824025818537
ML_positive12484807318094663318408868025571919516919459518691250
ML_negative1270556201783781991813951802301764911847151434195771334

Results of the Granger causality analysis for Apple trading volume variable

Granger causality
Time periodModelTwitter variableTwitter -> VolumeVolume -> Twitter
full periodVAR(5)Twitter volume**
VAR(6)negative VADER*
VAR(5)positive VADER*
first half of 2016VAR(2)negative VADER*
VAR(2)Twitter volume*
VAR(2)Fear*
VAR(3)negative NRC*
VAR (3)positive NRC*
VAR(2)neutral VADER*
VAR(2)positive VADER*
VAR(2)positive ML*
VAR(2)negative ML*
first half of 2017VAR(2)negative VADER**
VAR(1)negative NRC*
VAR(4)neutral VADER**
VAR(1)Anger*
VAR(1)positive VADER*
VAR(2)negative ML**
VAR(2)positive ML**

Descriptive statistics of affective indicators (2016–2017)

IndicatorMeanMinimumMaximumStandard DeviationCoefficient of Variation
Apple
Anger14,83012916,59111 %
positive (NRC)2,940564,37149 %
Fear20,96018824,62117 %
negative (NRC)54,02169256,11104 %
Sadness12,89042621,60167 %
negative (VADER)170,29101401,0157,9193 %
positive (VADER)353,52181789,0230,6665 %
neutral (VADER)558,32313771415,1274 %
negative machine learning499,15302593326,0765%
positive machine learning863,99354086495,2057%
Amazon
Anger5,020655,90118 %
positive (NRC)76,27078450,3867 %
Fear11,1801048,6577 %
negative (NRC)30,40028222,4974 %
Sadness6,220716,48104 %
negative (VADER)96,43080564,1066 %
positive (VADER)284,5002427177,2363 %
neutral (VADER)296,9201856143,8948 %
Positive machine learning453,8803531259,4357 %
Negative machine learning228,0701647132,9958 %

Information criteria AIC, BIC, and HQC for Amazon in the period 01/01/2016–31/12/2017

VariableAICBICHQC
volume - Twitter735
negative (VADER)734
fear (NRC)745
negative (NRC)735
positive (NRC)735
Sadness (NRC)747
neutral (VADER)523
positive (VADER)737
anger (NRC)957
positive (machine learning)735
negative (machine learning)735

Results of the Granger causality analysis for the Amazon trading volume variable

Granger causality
Time periodModelTwitter variableTwitter -> VolumeVolume -> Twitter
Full periodVAR(2)neutral VADER*
VAR(3)negative ML**
First half of 2016VAR(2)Twitter Volume*
VAR(2)Negative VADER*
VAR(2)Fear*
VAR(2)positive NRC*
VAR(2)neutral VADER*
VAR(2)positive VADER*
VAR(4)Anger*
VAR(2)negative ML*
VAR(2)negative ML*
First half of 2017VAR(3)Anger**

Information criteria AIC, BIC, and HQC for Apple in the period 01/01/2016–30/12/2017

VariableAICBICHQC
volume - Twitter844
negative (VADER)644
fear (NRC)737
negative (NRC)644
positive (NRC)746
Sadness (NRC)636
neutral (VADER)746
positive (VADER)746
anger (NRC)956
positive (machine learning)856
negative (machine learning)755

ADF and KPSS test results for trading volume and the set of Twitter-related explanatory variables for Amazon in the period 01/01/2016–31/12/2017

VariableADF test (H0: unit root exists, a=1; process is; proces I(1); α=5%)KPSS test (H0: the process is stationary)
AugmentationTest statisticp-valueAugmentationTest statistic
anger (NRC)Levels of the variable−3,654420,0048272,022180,462
First differences of the variable−9,538039,85e-180,008661980,462
fear (NRC)Levels of the variable−6,117816,205e-081,679690,462
First differences of the variable−14,24717,032e-330,01116780,462
negative (NRC)Levels of the variable−6,359631,558e-082,123380,462
First differences of the variable−9,10272,291e-160,01125550,462
positive (NRC)Levels of the variable−6,934934,954e-102,311230,462
First differences of the variable−10,05922,144e-190,008568020,462
Sadness (NRC)Levels of the variable−7,349493,634e-111,825550,462
First differences of the variable−8,821861,694e-150,00974750,462
volume - TwitterLevels of the variable−7,265476,221e-111,481810,462
First differences of the variable−10,04022,467e-190,009218940,462
volumeLevels of the variable−7,952356,891e-130,9323060,462
First differences of the variable−10,62113,279e-210,02375380,462
negative (VADER)Levels of the variable−7,05152,401e-101,07820,462
First differences of the variable−9,299765,549e-170,009357090,462
neutral (VADER)Levels of the variable−10,12641,304e-190,2338140,462
First differences of the variable
positive (VADER)Levels of the variable−6,57584,375e-092,668920,462
First differences of the variable−10,14421,143e-190,008960290,462
positive (machine learning)Levels of the variable−9,585926,946e-181,360720,462
First differences of the variable−10,03512,562e-190,008838010,462
negative (machine learning)Levels of the variable−5,904142,03e-071,542540,462
First differences of the variable−9,686593,328e-180,01109510,462

MAE results for the Apple and Amazon volume from VAR and LSTM approaches (test dataset)

Volume AppleVolume Amazon
first half of 2016
VARLSTMVARLSTM
basic model 52084210,079 30415580,280
Fear647500043234472,137171000029134831,193
Anger 44926837,853233390027106026,803
negative VADER 48940743,521130240028507242,885
negative NRC1166900047438402,462 27761003,405
neutral VADER793570047392150,78311019031538788,185
positive NRC527240046575716,841177150030197920,410
positive VADER 46123079,24821989031073486,591
Sadness 45630371,911 27609196,773
Twitter volume789450045452760,479282190029480840,315
ML positive506570048447840,46520198028161706,284
ML negative648020045159658,72174824029443364,882
first half of 2017
basic model 29959620,193 19816644,131
Fear 30032320,014 19610157,993
Anger2760600029989363,986940990019193604,890
negative VADER2906100029164579,531 19828186,097
negative NRC2622800029378243,531 19651018,103
Neutral VADER2872200029543515,048 19624056,441
Positive NRC2743700029435291,269 19873546,559
positive VADER2988000029613845,434 19708110,731
Sadness 29202756,331 19887700,586
Twitter volume2779500029200650,234 19666939,683
ML positive2890500029288276,166 19862915,138
ML negative29480424,66220186129,690
full period
basic model 37060274,722 49244981,060
Fear 36283377,928 22495202,131
Anger 37418912,444 22924401,013
negative VADER 37068676,291 22832473,868
negative NRC 37277158,450 22121621,013
Neutral VADER 36841175,338860350023183559,170
Positive NRC 37187736,091 22731837,658
positive VADER670350038351343,147 22735923,064
Sadness 36132565,891 22154059,560
Twitter volume 36132565,891 22725973,268
ML positive 36672397,669 22310880,027
ML negative 35741492,741970850022808249,226

ADF and KPSS test results for trading volume and the set of Twitter-related explanatory variables for Apple in the period 01/01/2016–31/12/2017

VariableADF test (H0: unit root exists, a=1; process is; proces I(1); α=5%)KPSS test (H0: the process is stationary)
AugmentationTest statisticAugmentationTest statisticAugmentation
anger (NRC)Levels of the variable−3,276120,016013,006920,462
First differences of the variable−11,991,125e-250,01618470,462
fear (NRC)Levels of the variable−3,979060,0015282,545750,462
First differences of the variable−11,35971,285e-230,01795490,462
negative (NRC)Levels of the variable−4,128560,00086772,181290,462
First differences of the variable−8,086762,779e-130,01467010,462
positive (NRC)Levels of the variable−5,918421,878e-071,808310,462
First differences of the variable−8,466022,054e-140,01061070,462
Sadness (NRC)Levels of the variable−5,451642,222e-061,151780,462
First differences of the variable−10,72831,47e-210,01041960,462
volume - TwitterLevels of the variable5,401842,727e-050,1760540,462
First differences of the variable−9,627495,127e-180,01347420,462
volumeLevels of the variable−5,260135,816e-062,541960,462
First differences of the variable−8,234071,018e-130,008083360,462
negative (VADER)Levels of the variable−3,849410,0024523,243970,462
First differences of the variable−11,09659,295e-230,01721410,462
neutral (VADER)Levels of the variable−5,289125,038e-062,747170,462
First differences of the variable−3,162722,13e-180,01603220,452
positive (VADER)Levels of the variable−6,428071,046e-080,9764740,462
First differences of the variable−11,48485,016e-240,01204370,462
positive (machine learning)Levels of the variable−5,875522,374e-072,071790,462
First differences of the variable−9,841911,064e-180,008139730,462
negative (machine learning)Levels of the variable5,707e-06−5,263962,952980,462
First differences of the variable−11,09919,114e-230,008503370,462

Pearson linear correlation coefficients for sentiment across different research periods (variable: volume)

IndicatorAAPLAMZNAAPLAMZNAAPLAMZN
full periodfirst half of 2016first half of 2017
critical value (5%, two tail)0,0860,0860,1720,1730,1740,174
Fear0,220,380,000,440,300,54
Anger0,210,000,110,050,300,15
negative VADER0,570,500,380,460,500,71
negative NRC0,490,510,310,650,400,73
neutral VADER0,560,570,290,410,500,77
positive NRC0,400,300,260,290,400,59
positive VADER0,560,440,360,460,600,70
Sadness0,290,400,370,530,300,52
Volume0,540,520,360,460,500,74
Machine learning_positive0,530,480,140,380,400,72
Machine learning_negative0,520,550,160,530,300,74
DOI: https://doi.org/10.2478/ceej-2025-0024 | Journal eISSN: 2543-6821 | Journal ISSN: 2544-9001
Journal RSS Feed
Language: English
Page range: 412 - 432
Submitted on: Nov 22, 2024
|
Accepted on: Nov 5, 2025
|
Published on: Dec 29, 2025
Published by: Faculty of Economic Sciences, University of Warsaw
In partnership with: Paradigm Publishing Services
Publication frequency: 1 issue per year
Keywords:
behavioural economics,
VAR,
LSTM,
sentiment analysis,
NLP
Related subjects:
Business and economics,
Political economics,
Economic theory, systems and structures,
Microeconomics,
Macroeconomics,
Economic policy

© 2025 Joanna Michalak, published by Faculty of Economic Sciences, University of Warsaw
This work is licensed under the Creative Commons Attribution 4.0 License.

Previous articleVolume 12 (2025): Issue 59 (January 2025)