









































Pa
ge

 
1



Pa
ge

 
70

American Journal of  Smart 
Technology and Solutions (AJSTS)

R-Shiny Web Application Development for Multilayer Perceptron State Switching Model 
for Predicting Regimes of  Time Series Returns   

Oluwasegun Agbailu Adejumo1*, Omorogbe Joseph Asemota1, Samuel Olayemi Olanrewaju1

Volume 4 Issue 2, Year 2025
ISSN: 2837-0295 (Online)

DOI: https://doi.org/10.54536/ajsts.v4i2.5956
https://journals.e-palli.com/home/index.php/ajsts

Article Information ABSTRACT

Received: August 14, 2025

Accepted: September 22, 2025

Published: October 21, 2025

Forecasting regimes in financial time series is complicated by nonlinearity and small-
sample limitations, where conventional Markov Switching Models (MSMs) and regime-
switching autoregressive models often underperform. To address this, we developed an 
interactive R-Shiny Web Application implementing the recently introduced Multilayer 
Perceptron State Switching Model (MLPSSM). The app integrates neural networks to 
capture nonlinear intra-regime patterns with Markov switching structures to identify 
latent regime transitions. Using Nigerian exchange rate returns as a case study, the app 
demonstrated robust performance across diagnostics, estimation, and forecasting. Residual 
checks confirmed that the hybrid approach effectively modeled underlying dynamics, 
while forecasts achieved lower RMSE and MAE than baseline MSMs. The web-based 
interface further enhances accessibility, enabling both technical and non-specialist users 
to apply advanced regime-switching methods without coding expertise. The MLPSSM 
Web App thus bridges machine learning and econometric modeling, offering a practical, 
reproducible tool for regime prediction in financial markets. 

Keywords
Financial Time Series Returns, 
Markov Process, MLPSSM, R 
Shiny, Regime Switching

1 Department of  Statistics, Faculty of  Science, University of  Abuja, Abuja, Nigeria
* Corresponding author’s e-mail: agbailuoa@gmail.com

INTRODUCTION 
Financial time series data, such as stock returns, exchange 
rates, and commodity prices, are often characterized by 
nonlinearity, volatility clustering, and structural regime 
shifts. Traditional linear time series models, while useful 
for short-term forecasting, are limited in their ability to 
capture abrupt changes between economic states such 
as bull and bear markets or high- and low-volatility 
regimes (Hamilton, 1989; Krolzig, 1997). To address 
these limitations, researchers have developed state-
switching models, including the Markov Switching Model 
(MSM) and Regime-Switching Autoregressive Models, 
which explicitly account for stochastic regime shifts and 
structural breaks. These models have been widely applied 
in empirical finance to enhance the understanding of  
market dynamics and improve predictive performance in 
risk-sensitive environments (Ang & Timmermann, 2012).
Despite their popularity, conventional MSMs and 
regime-switching autoregressive models demonstrate 
poor predictive performance in small-sample settings, 
which is a common challenge in emerging markets and 
in high-frequency but short-span datasets. Under such 
conditions, parameter estimates become unstable, leading 
to unreliable regime classification, weak predictive power, 
and difficulty in distinguishing genuine structural shifts 
from random noise (Psaradakis & Spagnolo, 2003; 
Chauvet & Potter, 2000). These shortcomings limit the 
applicability of  MSMs and related models, especially in 
contexts where robust regime detection is most critical 
for investors, policymakers, and risk managers.
To overcome these challenges, recent research has 
explored the integration of  neural networks into regime-

switching frameworks. Neural networks excel at capturing 
nonlinear patterns, hidden structures, and complex 
dependencies within financial data that conventional 
linear models often fail to detect (Zhang et al., 1998). 
By combining the flexibility of  machine learning with 
the interpretability of  regime-based approaches, Neural 
Network State Switching Models (NN-SSMs) have 
emerged as promising alternatives for predicting time 
series regimes (Yao et al., 1999).
A notable advancement in this field is the study by 
Adejumo et al. (2025), who introduced the Multilayer 
Perceptron State Switching Model (MLPSSM), a 
novel hybrid approach that embeds a neural network 
within a regime-switching framework. Their findings 
demonstrated that the MLPSSM consistently outperforms 
traditional MSMs and regime-switching autoregressive 
models, particularly in small-sample environments and 
in accurately predicting regime transitions. By addressing 
the weaknesses of  conventional models, the MLPSSM 
provides a more robust and reliable tool for forecasting 
financial returns under volatile and structurally shifting 
conditions.
However, while Adejumo et al. (2025) established the 
methodological superiority of  the MLPSSM, its practical 
application remains constrained by accessibility barriers. 
Financial analysts and policymakers often lack the 
programming expertise to implement advanced hybrid 
models, and existing statistical software environments do 
not provide user-friendly, interactive platforms for real-
time regime prediction.
In response, the present study focuses on the development 
of  an R-Shiny Web Application for the MLPSSM 



Pa
ge

 
71

https://journals.e-palli.com/home/index.php/ajsts

Am. J. Smart. Technol. Solutions 4(2) 70-79, 2025

introduced by Adejumo et al. (2025). R-Shiny provides 
an ideal platform for operationalizing this optimal model 
into an interactive, scalable, and accessible tool. Through 
this web application, users will be able to upload datasets, 
configure model parameters, visualize regime-switching 
dynamics, and generate real-time forecasts without the 
need for advanced coding. Thus, this research makes a 
dual contribution: methodologically, by extending the 
work of  Adejumo et al. (2025) through application-
oriented development of  the MLPSSM; and practically, 
by democratizing access to advanced predictive models 
for financial analysts, researchers, and policymakers in 
Nigeria and other emerging economies where regime 
shifts and structural breaks are frequent.

MATERIALS AND METHODS
Following Hamilton (2005), the MSSM addresses the 
hidden states weakness of  the Hidden Markov Model 
(HMM). To describe the MSSM, we assume that the 
number of  states (or States) is N=2, i.e. St  Ω={1,2}. 
This implies that; for instance, the log returns of  financial 
time series are drawn from distinct normal distributions, 
depending on what state the HMM is currently in. This 
would give us the following model to work with:

This means that when the state of  the HMM for time 
t is 1, then the expectation of  the dependent variable is 
μ1 and the variance of  the innovations is σ1

2, similarly 
when the state of  the HMM for time t is 2, then the 
expectation of  the dependent variable is μ2 and the 
variance of  the innovations is σ2

2 and so on. Since the 
underlying Markov chain is hidden one cannot observe 
what state the HMM is in directly, but only deduce its 
operation through the observed behaviour of  Yt. In order 
to attain the probability law governing the observed data 
Yt a probabilistic model of  what causes the change from 
state St=i to state St=j is required. This can be specified 
using the transition probabilities of  an N=2 state HMM;
ρi,j=Pr (St=j│St-1=i)    i,j∈Ω={1,2}           ....(2)
The transition probability (2) is by the Markov property; 
dependent of  the past only through the value of  the 
most recent state. This is one of  the central points of  
the structure of  a Markov state switching model, i.e. 
the switching of  the states of  the underlying HMM is a 
stochastic process itself.
There are several ways to estimate the required parameters 
of  the 2-state MSSM given by (1), for instance by using 
the maximum likelihood estimation method.
The Multilayer Perceptrons State Switching Model 
(MLPSSM) by Adejumo et al. (2025)
To improve the regime prediction performance of  the 
famous state switching model described in equation 
(1) for nonlinear time series data, especially returns, 
in small sample size contexts, Adejumo-Agbailu et al. 
(2025) integrate deep neural networks such as Multilayer 
Perceptrons (MLP) with Markov two-State Switching 

modelling technique. The Adejumo-Agbailu et al. (2025) 
neural network state switching modeling approach 
combines the strengths of  MLP-deep neural networks 
(i.e., efficient in high complexity datasets) and Markov 
two-state switching capabilities of  ensuring both 
interpretability and predictability of  the two Nigerian 
exchange returns’ regimes (i.e., bull and bear states). The 
models were introduced in two phases.

1st Phase: Generative Network using MLP
Given time series dataset of  Yt, the generative network 
procedure for the dataset include the following: 
Define of  training and testing dataset of  Yk i.e. by setting 
training dataset at time step tk, and testing dataset at time 
step tt-k
At time step tk, MLP is used to process the input data (i.e. 
training dataset) such as hk~mlph (yt ).

2nd Phase: Integration of  Generative Network ht into 
the Markov Two-State Switching Model 
Subsequently, the generative network ht is integrated in the 
Markov two-state switching model alongside the defined 
training dataset to develop the Multilayer Perceptrons 
State Switching Model (MLPSSM). This would give us 
the following model to work with:
Ytk~ht)=gt=cst+Β1(gt-1-cst-1)+Β2 (gt-2-cst-2))+ϵt            ....(7)
where
cst=c0 S0t+c1 S1t+c2 S2t  
σst

2=σ1
2 S1t+σ2

2 S2t
ϵt~i.i.d(0,σst

2)
cst is the state dependent mean, σst

2 is state dependent 
variance and the coefficients are Β1or Β2; which could be 
different for different subsamples. The proposal will be 
to model the state St as the outcome of  an unobserved 
two-state Markov chain with St independent of  ϵt for all t.
The transitions of  the St, are presumed to be ergodic and 
intricate first order Markov-process. This means impacts 
of  earlier observation(s) for the gt and state(s) is/are 
completely captured in the recent gt state(s) observations 
as represented in (3); 
ρij=Prob(St=j ⁄ St-1) =i)    ∀i,j=1,2∑i=1

2ρij =1          ....(8)
Matrix P captures the probability of  switching which is 
known as a transition matrix;
              ....(9)

    
Where, P11+P21=1,  and P12+P22=1 
The nearer the probability ρij is to one the longer it takes 
to shift to the next regime.
Consider the model given by equation (7), i.e., a Markov 
regime-switching model with 2-regimes. The estimation 
will be performed using Hamilton’s filter, where the 
main idea is to calculate each state’s filter probabilities by 
making inferences on each state’s unknown probabilities 
based on the available information. 

The MLPSSM Web Application Development
MLPSSM is a web application developed in the Shiny 
environment, leveraging a variety of  R libraries and 



Pa
ge

 
72

https://journals.e-palli.com/home/index.php/ajsts

Am. J. Smart. Technol. Solutions 4(2) 70-79, 2025

packages to provide a comprehensive regimes of  time 
series returns modelling and predicting experience. 
Currently, there are more than 12,000 R packages or 
libraries available, which are the result of  a collaborative 
project sustained by individuals from different parts 
of  the world and disciplines. The project continues 
to grow both in the number of  packages and in 
knowledge areas, such as statistics, finance, genetics, 
network analysis, and data mining, among many others 
(Cruz et al. 2023). This application has been designed to 
facilitate efficient and effective time series returns data 
exploration, preliminary analysis, regime modelling and 
predicting.
To ensure optimal functionality and a wide range of  
features, MLPSSM makes use of  libraries and packages 
in R.

You can view the application at the following URL: 
https://agbailuoa.shinyapps.io/MLPSSM_App/ 
The user interface of  MLPSSM has been carefully 
designed using the Shiny package in R, providing a 
smooth and intuitive experience for users. Shiny is an R 
package that allows the construction of  interactive web 
applications from R scripts. The application is organized 
into different tabs, each with specific functionalities.
The application architecture includes:

i. Frontend; shiny, shinydashboard for UI (upload, 
plots, model outputs).

ii. Backend; Neural networks via RSNNS::mlp, regime-
switching via MSwM::msmFit.

iii. Support packages; forecast, tseries, lmtest, Metrics, 
ggplot2 for tests, diagnostics, and forecasting.
Figure 1 presents a flowchart of  the MLPSSM App.

Figure 1: Flowchart of  the Application: MLPSSM App

MLPSSM App User Interface
The user interface of  MLPSSM App welcomes you to an 
interactive and enriching experience. Here you will find 
a concise description of  the application’s functionalities, 
Figure 2 presents the user interface of  the developed 
MLPSSM App. 
According to Figure 2, the application includes tabs for 
Upload Data, a Time Series, a Model Output, Residuals, 
and a Forecast. The Upload Data tab allows users to load 
returns time series, i.e. rt=ln(Rt⁄Rt-1 ), in CSV format. The 
Time Series tab presents users with the time series plot 
and preliminary tests of  the series such as the ADF test 
(stationarity test) and BDS test (non-linearity test). The 
Model Estimation tab allows users to estimate the novel 
MLPSSM and provides the estimated model’s transition 
probability plots. The Residuals tab allows users to assess 
the diagnosed residuals of  the estimated MLPSSM, 
including the residuals plot, Ljung-Box test, and Shapiro-
Wilk test. Lastly, the Forecast tab enables users to forecast 
or predict the returns-regime means and the model 
performance metrics

Mlpssm App Evaluation
This section presents and discusses the deployment and 

evaluation of  the MLPSSM Web App using the sample 
size 30 of  daily Nigeria Exchange Rate returns. Figure 
3 presents the “Data Upload” interface of  the app 
using the aforementioned dataset. The app successfully 
ingested the Nigeria Exchange Rate (USD/₦) daily data 
from CBN (11th June–23rd July 2025). The data upload 
interface was limited to CSV files, ensuring standardized 
inputs. The app extracted the second column of  the 
dataset as the time series and visualized it under the Time 
Series tab. This confirms the reactivity of  the app, as the 
uploaded data was immediately transformed into plots 
and summary previews for the user. By visualizing the 
exchange rate returns, the app provides users with an 
intuitive entry point before model estimation.
Subsequently, the time series is visualized in the “Time 
Series” tab, just below the “Upload & Settings” tab. The 
plotOutput function is used to display the graph of  the 
time series in levels. The reactivity of  the application is 
reflected in the outputs, which are the results (numeric 
values, plots) received by the interface from the server.R. 
In our case, the result is a graph and it is inserted using 
the plotOutput() function (Figure 4).
The inclusion of  Augmented Dickey-Fuller (ADF) and 
BDS (Brock-Dechert-Scheinkman) tests within the Time 



Pa
ge

 
73

https://journals.e-palli.com/home/index.php/ajsts

Am. J. Smart. Technol. Solutions 4(2) 70-79, 2025

Figure 2: The User Interface of  MLPSSM Web App 

Figure 3: Data Upload Interface of  the MLPSSM Web App  

Figure 4: Time Series Plot of  the Nigeria Exchange Rate Returns using the Time Series tab of  the App  

Series tab allowed for formal pre-model diagnostics.
- ADF Test: Evaluates whether the returns are 

stationary. If  the null hypothesis of  a unit root is rejected, 
the series can be considered stationary. In most exchange 
rate returns, stationarity is expected after transformation 
into returns.

- BDS Test: Evaluates non-linearity and independence 

of  residuals. Significant results indicate the presence of  
hidden structure, justifying the need for nonlinear hybrid 
models like MLPSSM.
In this evaluation, the results confirmed stationarity and 
non-linearity, validating the use of  a nonlinear, regime-
switching hybrid model.



Pa
ge

 
74

https://journals.e-palli.com/home/index.php/ajsts

Am. J. Smart. Technol. Solutions 4(2) 70-79, 2025

Figure 5: ADF Test Results of  the Nigeria Exchange Rate Returns using the Time Series tab 

Figure 6: BDS Test Results of  the Nigeria Exchange Rate Returns using the Time Series tab 



Pa
ge

 
75

https://journals.e-palli.com/home/index.php/ajsts

Am. J. Smart. Technol. Solutions 4(2) 70-79, 2025

Figure 7: MLPSSM Estimation for the Nigeria Exchange Rate Returns using the Model Estimation tab  

Figure 8: Regimes Transition Probabilities of  MLPSSM Estimated of  the Nigeria Exchange Rate Returns using the 
Model Estimation tab

The Model Estimation tab provided the estimated 
parameters of  the MLPSSM (a combination of  neural 
networks for nonlinear fitting and Markov switching 
models for regime identification).

- Estimation Results (Figure 7): These show the fitted 
coefficients and transition probabilities of  the model. 
The estimates suggest the model successfully learned 
from the exchange rate returns.

- Transition Probabilities (Figure 8): The graph 

displayed probabilities of  switching between regimes 
(bear vs. bull). High self-transition probabilities suggest 
persistence within regimes, while non-trivial switching 
probabilities capture regime dynamics.
This indicates that the app correctly implements the 
regime identification logic, a major improvement over 
conventional MSMs and ARSSMs, which often perform 
poorly in small samples.



Pa
ge

 
76

https://journals.e-palli.com/home/index.php/ajsts

Am. J. Smart. Technol. Solutions 4(2) 70-79, 2025

Figure 9: Residuals Plot of  the Estimated MLPSSM using the Residuals tab

Figure 10: Residuals Diagnostics of  the Estimated MLPSSM using the Residuals tab



Pa
ge

 
77

https://journals.e-palli.com/home/index.php/ajsts

Am. J. Smart. Technol. Solutions 4(2) 70-79, 2025

Figure 11: Forecast Plot of  the Estimated MLPSSM using the Forecast tab 

Figure 12: Forecast Estimate of  the Estimated MLPSSM using the Forecast tab  



Pa
ge

 
78

https://journals.e-palli.com/home/index.php/ajsts

Am. J. Smart. Technol. Solutions 4(2) 70-79, 2025

Figure 13: Estimated MLPSSM Performance Metrics using the Forecast tab 

Afterward, The Residuals tab provided three diagnostic 
checks: residual plots, Ljung-Box test, and Shapiro-Wilk test.

- Residual Plot (Figure 9): Shows whether errors 
fluctuate randomly around zero. A fairly random pattern 
indicates good fit.

- Ljung-Box Test (Figure 10): Checks if  residuals are 
autocorrelated. A non-significant result suggests residual 
independence.

- Shapiro-Wilk Test (Figure 10): Tests normality of  
residuals. This ensures validity of  statistical inferences.
Together, these diagnostics confirm whether the hybrid 
model adequately captures the dynamics in the data. In 
this evaluation, the residual diagnostics suggested that 
the MLPSSM provides a statistically adequate fit, with 
minimal autocorrelation and approximate normality.
The Forecast tab generated three outputs:

- Forecast Plot (Figure 11): Displayed short-horizon 
projections of  exchange rate returns.

- Forecast Estimates with Regimes (Figure 12): Highlighted 
regime-dependent forecasts (bear vs. bull), making it useful 
for risk management and trading strategy design.

- Performance Metrics (Figure 13): Reported forecast 
accuracy measures (RMSE, MAE). These quantitative 
metrics confirmed the predictive strength of  the 
MLPSSM in small-sample forecasting.
The hybrid model outperformed conventional regime-
switching models in terms of  both forecast precision and 
regime classification accuracy, aligning with Adejumo et 
al. (2025)’s theoretical contributions.
The deployment and evaluation of  the MLPSSM Web 
App demonstrated, technical correctness (i.e. proper 
integration of  reactivity, diagnostics, regime estimation, 
and forecasting), practical relevance (i.e. useful 
outputs for researchers, policymakers, and investors 
in financial markets), methodological novelty (i.e. the 
app operationalizes Adejumo et al. (2025)’s neural-
network-based hybrid model, overcoming weaknesses of  
traditional MSMs in small samples), and user accessibility 
(i.e. with simple tabs and automated tests, non-technical 
users can upload data, check assumptions, run hybrid 
models, and obtain forecasts).

Discussion 
The evaluation of  the Multilayer Perceptron State Switching 
Model (MLPSSM) Web Application demonstrates its 
capacity to address long-standing methodological and 
practical challenges in regime prediction within financial 

time series. Conventional Markov Switching Models 
(MSMs) and regime-switching autoregressive models 
(ARSSMs) have often been limited in their predictive 
capacity, especially under small-sample conditions, where 
parameter instability and regime misclassification become 
prominent issues (Hamilton, 1989; Krolzig, 1997). Recent 
studies, including Adejumo et al. (2025), have argued that 
hybrid frameworks combining neural networks with 
regime-switching methods offer a robust alternative by 
integrating nonlinear approximation with probabilistic 
regime identification.
The performance of  the app across data input, model 
estimation, residual diagnostics, and forecasting validates 
these theoretical expectations. Pre-model diagnostics 
confirmed both stationarity and non-linearity in the 
Nigerian exchange rate returns, justifying the application of  
a nonlinear, regime-dependent framework. The MLPSSM 
successfully captured hidden regime dynamics through 
a two-stage estimation; first modeling nonlinear intra-
regime patterns using multilayer perceptrons (MLPs), and 
subsequently applying a Markov-switching structure to the 
residuals. The transition probability estimates highlighted 
persistent regime behaviors, while allowing for realistic 
switching, aligning with empirical evidence in financial 
markets where volatility clustering and regime persistence 
are common (Hamilton & Susmel, 1994). 
Residual diagnostics further reinforced the adequacy of  the 
hybrid approach. The Ljung-Box and Shapiro-Wilk results 
indicated that the residuals were largely uncorrelated and 
approximately normal, suggesting that the hybrid model 
extracted most of  the signal embedded in the series. These 
results compare favorably to conventional MSMs, which 
often leave strong autocorrelation structures unmodeled, 
thereby undermining their predictive reliability (Maheu & 
McCurdy, 2000).
The forecasting module provided perhaps the most 
critical evidence of  the app’s utility. Both the point 
forecasts and the regime-dependent projections showed 
strong predictive accuracy, as reflected in the RMSE and 
MAE metrics. These results substantiate the claim by 
Adejumo et al. (2025) that the MLPSSM can outperform 
baseline regime-switching methods, particularly in small 
samples where neural networks are able to flexibly 
capture nonlinear structures that traditional parametric 
forms miss. The integration of  forecasts with regime 
classifications also provides added interpretive value for 
practitioners in risk management and policy analysis, who 



Pa
ge

 
79

https://journals.e-palli.com/home/index.php/ajsts

Am. J. Smart. Technol. Solutions 4(2) 70-79, 2025

require not just point estimates but also insights into 
underlying market conditions.
Beyond methodological contributions, the deployment 
of  this model in a Shiny Web Application represents an 
important step toward accessibility and reproducibility. 
By embedding advanced methods in an interactive 
interface, the app democratizes the use of  sophisticated 
econometric techniques, enabling both technical and 
non-technical users to conduct diagnostics, fit models, 
and generate forecasts. This aligns with recent calls in 
computational economics and data science for tools that 
bridge methodological rigor and usability (Chambers & 
Hastie, 1992; Varian, 2014).

CONCLUSIONS
In summary, the MLPSSM Web App provides empirical 
evidence for the superiority of  neural-network-enhanced 
regime-switching approaches over conventional methods. 
It not only validates the theoretical advances proposed 
by Adejumo et al. (2025) but also operationalizes them 
in a user-friendly platform that can be applied to diverse 
time series forecasting tasks. These findings highlight 
both the academic novelty and the practical relevance 
of  integrating machine learning with classical state-space 
and regime-switching frameworks.
From a policy and practical perspective, the implications 
are significant. For policymakers in Nigeria and other 
emerging economies, the ability to identify and forecast 
financial regimes in exchange rate markets provides a 
valuable tool for anticipating volatility and implementing 
timely interventions. Regulators can apply the app’s 
diagnostic and forecasting features to monitor systemic 
risks, while financial institutions can leverage the regime 
predictions to inform hedging strategies and risk-adjusted 
investment decisions.
Moreover, by operationalizing advanced econometric and 
machine learning techniques within a Shiny Web App, this 
study bridges the gap between methodological innovation 
and practical accessibility. The interactive dashboard 
ensures that both technical experts and non-specialist 
users can upload data, conduct diagnostics, fit models, 
and interpret results without extensive coding skills. 
This democratization of  advanced forecasting methods 
supports capacity building within financial institutions, 
academic research, and policy circles.
Nevertheless, limitations remain. The two-stage 
estimation approach assumes that regime structure 
lies entirely in residual dynamics, which may overlook 
deeper joint interactions. Additionally, small-sample 
robustness, while improved over MSMs, still warrants 
further exploration through regularization techniques and 
Bayesian extensions. Future research should also extend 
the Web App to multivariate series and high-frequency 
data, as well as integrate real-time data streaming for 
dynamic monitoring.
In conclusion, the MLPSSM Web App provides a 

practical and innovative tool for regime prediction 
and forecasting in financial time series. It validates the 
theoretical contributions of  Adejumo et al. (2025) and 
extends them into a usable platform with direct policy 
and market relevance. The study contributes both to the 
advancement of  econometric methodology and to the 
provision of  applied solutions for managing uncertainty 
in financial markets.

REFERENCES 
Adejumo, O. A., Asemota, O. J. & Olanrewaju, S. O. (2025). 

Novel Neural Network State Switching Models for 
Returns Predicting with Regime Switching: A Monte 
Carlo’s Simulation Approach. American Journal of  
Applied Statistics and Economics, 4(1).

Ang, A., & Timmermann, A. (2012). Regime changes and 
financial markets. Annual Review of  Financial Economics, 
4, 313–337.

Chambers, J. M., & Hastie, T. J. (1992). Statistical models in 
S. Wadsworth & Brooks/Cole.

Chauvet, M., & Potter, S. (2000). Coincident and leading 
indicators of  the stock market. Journal of  Empirical 
Finance, 7(1), 87–111.

Cruz, T., Jimenez, F. G., Bravo, A. R. Q., & Ander, E. 
(2023). DataXploreFines: Generalized Data for Informed 
Decision, Making, An Interactive Shiny Application 
for Data Analysis and Visualization. arXiv preprint 
arXiv:2307.11056.

Hamilton, J. D. (1989). A new approach to the economic 
analysis of  nonstationary time series and the business 
cycle. Econometrica, 57(2), 357–384. https://doi.
org/10.2307/1912559

Hamilton, J. D., & Susmel, R. (1994). Autoregressive 
conditional heteroskedasticity and changes in regime. 
Journal of  Econometrics, 64(1–2), 307–333. https://doi.
org/10.1016/0304-4076(94)90067-1

Krolzig, H.-M. (1997). Markov-switching vector autoregressions: 
Modelling, statistical inference, and application to business cycle 
analysis. Springer.

Maheu, J. M., & McCurdy, T. H. (2000). Identifying bull 
and bear markets in stock returns. Journal of  Business 
& Economic Statistics, 18(1), 100–112. https://doi.org/
10.1080/07350015.2000.10524846

Psaradakis, Z., & Spagnolo, N. (2003). On the 
determination of  the number of  regimes in Markov-
switching autoregressive models. Journal of  Time Series 
Analysis, 24(2), 237–252.

Varian, H. R. (2014). Big data: New tricks for econometrics. 
Journal of  Economic Perspectives, 28(2), 3–28. https://
doi.org/10.1257/jep.28.2.3

Yao, J., Tan, C. L., & Poh, H. L. (1999). Neural networks 
for technical analysis: A study on KLCI. International 
Journal of  Theoretical and Applied Finance, 2(2), 221–241.

Zhang, G., Patuwo, B. E., & Hu, M. Y. (1998). Forecasting 
with artificial neural networks: The state of  the art. 
International Journal of  Forecasting, 14(1), 35–62.


