Forecasting Stock Prices of Fintech Companies of India using Random Forest with High-frequency Data

Abstract

The fintech segment is currently one of the most rapidly growing industries, attracting numerous investors who anticipate substantial returns in the future. Notably, not only individual retail investors but also mutual fund agencies are actively engaged in predicting stock prices within this sector to maximize their trading gains. The purpose of the study is to formulate stock forecasting models for top three Fintech Companies of India i.e., Policy Bazar, One 97 Communications Paytm Ltd., and Niyogin Ltd. Using Random Forest model with high-frequency data in Python. The literature review section also proves that this study is a novel piece of work as none of the existing research study focused on predicting stock prices of Fintech Companies of India using Random Forest model. The data is extracted from www.moneycontrol.com and www.kotaksecurities.com, for the period from 1st October, 2022 – 30th September, 2023. The study deals about 293,280 data points i.e., 3 companies @ 97,760 each. It has been found that the forecasting model of random forest provides very successful results for prediction as the co-efficient of determination of all the selected companies is more than 95%.

Full text

Journal of Open Innovation: Technology, Market, and Complexity 10 (2024) 100180
Available online 18 November 2023
2199-8531/© 2023 Published by Elsevier Ltd on behalf of Prof JinHyo Joseph Yun. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Forecasting stock prices of ntech companies of India using random forest
with high-frequency data
Bharat Kumar Meher
a
, Manohar Singh
b
, Ramona Birau
c
,
*
, Abhishek Anand
d
a
PG Department of Commerce & Management, Purnea University, Purnea, Bihar 854301, India
b
Department of Commerce, Government Autonomous PG College, Chhindwara, Madhya Pradesh, India
c
Faculty of Economic Science, University Constantin Brancusi of Tg-Jiu, Romania
d
PG Department of Economics, Purnea University, Purnea, Bihar 854301, India
ARTICLE INFO
JEL classication:
C55
G17
G23
Keywords:
Forecasting
Random Forest
Fintech
High-frequency data
Python
ABSTRACT
The ntech segment is currently one of the most rapidly growing industries, attracting numerous investors who
anticipate substantial returns in the future. Notably, not only individual retail investors but also mutual fund
agencies are actively engaged in predicting stock prices within this sector to maximize their trading gains. The
purpose of the study is to formulate stock forecasting models for top three Fintech Companies of India i.e., Policy
Bazar, One 97 Communications Paytm Ltd., and Niyogin Ltd. Using Random Forest model with high-frequency
data in Python. The literature review section also proves that this study is a novel piece of work as none of the
existing research study focused on predicting stock prices of Fintech Companies of India using Random Forest
model. The data is extracted from www.moneycontrol.com and www.kotaksecurities.com, for the period from
1st October, 2022–30 th September, 2023. The study deals about 293,280 data points i.e., 3 companies @ 97,760
each. It has been found that the forecasting model of random forest provides very successful results for prediction
as the co-efcient of determination of all the selected companies is more than 95%.
1. Introduction
Technological advancements have a longstanding presence in the
nancial sector. Digital innovation, in particular, has ushered in sig-
nicant enhancements in system connectivity, computational capabil-
ities, cost efciency, and the generation of actionable data. These
improvements have led to the reduction of transaction costs and the
emergence of novel business models and players in the nancial land-
scape (Feyen et al., 2021). These new entrants are collectively referred
to as "Fintech." In the digital age, numerous Fintech startups have
proliferated, both in developed and developing nations, including India.
Fintech, as the name suggests, is the fusion of nance and technology. It
gained substantial momentum following the global nancial crisis of
2008 and continues to evolve rapidly, with many unexplored opportu-
nities remaining (Taujanskait˙
e and Kuizinait˙
e, 2022).
In this dynamic environment, many market participants leverage
technology to streamline nancial services, encompassing lending, in-
surance, investments, trading, budgeting, and more (Scardovi, 2017;
Pazarbasioglu et al., 2020). This contributes to the seamless and efcient
operation of nancial services traditionally offered by banks and
insurance companies (Alt et al., 2018; Breidbach et al., 2020). Fintech
companies in India, such as Paytm, gained prominence during events
like Demonetization and the COVID-19 pandemic when cashless trans-
actions became the preferred choice for many (Jakhiya et al., 2020;
Moid and Shankar, 2022; Khando et al., 2023). As the ntech sector
expands, numerous players in India are narrowing their focus to niche
areas. Consumer lending ntech rms constitute a signicant portion,
comprising 17% of the total ntech enterprises (Nenavath, 2022;
Migozzi et al., 2023). The demand for credit in India continues to rise,
prompting banks to collaborate with ntech companies to enhance
service offerings. For example, Paytm simplies payments by reducing
the need for manual intervention in card and net banking transactions
(Ganjoo et al., 2023). Angel investors are increasingly drawn to invest in
ntech startups, recognizing the industry’s substantial growth potential
(Surana et al., 2020; Harris, 2021; Saura et al., 2021). Some ntech
companies have even reached a point where they expand operations and
become publicly listed on stock exchanges.
The ntech sector is currently one of the most rapidly growing in-
dustries, attracting numerous investors who anticipate substantial
returns in the future (Mention, 2019; Zhang-Zhang et al., 2020; Arora
* Corresponding author.
E-mail addresses: ramona.f.birau@gmail.com (R. Birau), abhi2eco@gmail.com (A. Anand).
Contents lists available at ScienceDirect
Journal of Open Innovation: Technology, Market,
and Complexity
journal homepage: www.sciencedirect.com/journal/journal-of-open-innovation-technology-
market-and-complexity
https://doi.org/10.1016/j.joitmc.2023.100180
Received 15 October 2023; Received in revised form 8 November 2023; Accepted 12 November 2023

Journal of Open Innovation: Technology, Market, and Complexity 10 (2024) 100180
2
and Madan, 2023). Notably, not only individual retail investors but also
mutual fund agencies are actively engaged in predicting stock prices
within this sector to maximize their trading gains (Palmi´
e et al., 2020;
Bhatia et al., 2021). Various techniques, such as Exponential Moving
Averages (EMA), AutoRegressive Integrated Moving Average (ARIMA),
GARCH Models, and Holt-Winters Exponential Smoothing, have been
employed to forecast stock prices using univariate data (Priyamvada &
Wadhvani, 2017; Chatterjee et al., 2021). However, the Random Forest
method emerges as a promising alternative. Random Forest is an
ensemble learning approach that amalgamates multiple decision trees
for making predictions (Asad, 2015; Thakur and Kumar, 2018). This
ensemble strategy often yields more robust and accurate predictions
compared to individual models, with the added advantage of accom-
modating non-linearities. While comparing the Random Forest model
with the other benchmark model like ARIMA, random forest model can
also consider non-linear patterns in stock prices movement, and can
successfully handle seasonality and trends, outliers and anomalies and
both short-term and long-term forecasting also.Moreover, it is versatile
enough to handle both regression and classication tasks, making it
suitable for various stock price forecasting scenarios (Valencia et al.,
2019; Landis and Cha, 2020; Mohanta et al., 2020). In this study, we
endeavor to predict stock prices in the burgeoning Indian FinTech sector
using the Random Forest model, employing high-frequency data. The
use of high-frequency data, specically one-minute closing prices of
Indian ntech companies, enhances prediction accuracy and precision.
Innovation Dynamics research often explores how technological ad-
vancements impact various sectors, including nance. The use of such
kind of high-frequency data and machine learning techniques like
Random Forest to forecast stock prices of ntech companies suggests an
application of technological innovation in the nancial industry.Sub-
sequently, the literature review section outlines key research endeavors
related to stock price forecasting, high-frequency data, and Random
Forest. This study not only substantiates the research gap but also un-
derscores the novelty of our research within this eld.
2. Review of literature
India is well-known as a thriving center for ntech, and as the Indian
start-up ecosystem expands, more industries inspired by ntech use
cases will start up and receive funding from different sources. Risky
investments in the ntech segment of the equities market may pay off
handsomely, according to a study that indicates a signicant risk-return
link in the Indian ntech industry (Mention, 2019; Brown and Wiles,
2020; Bhatnagar et al., 2022).
Akyildirim et al. (2023) test many machine learning strategies for
their ability to foretell intraday excess returns. Prediction rates much
above 50% are generated by machine learning analytics, and optimal
prot ratios can go as high as 33%. The ndings support the usefulness
of analytics and machine learning techniques and prompt additional
discourse on the market’s moderate efciency. Data from the Indian
stock market shows that Levenberg-Marquardt (LM), Scaled Conjugate
Gradient, and Bayesian regularization algorithms all achieve an accu-
racy of 99.9% when using tick data (Selvamuthu et al., 2019). When
compared to the ndings obtained using tick data, the accuracy over a
15-minute dataset lowers to 96.2% for LM, 97.0% for SCG, and 98.9%
for Bayesian Regularization.
The potential gains from accurate future forecasting have made it a
goal of many societies and economies. With the help of AI, scientists will
have access to more precise predictions than ever before. Over time, as
technology and algorithms improve, they will become more precise.
Overall, feature engineering proved to benet the models (Alkhatib
et al., 2022). When applied to models using Long Short-Term Memory,
the new method yielded signicant improvements. Predicting the Fin-
tech index is useful for a number of different people since it can help
investors create successful short-, medium-, and long-term investment
plans and can direct nancial regulators toward making accurate and
effective regulatory rules. These ndings show that the algorithm can be
used as a more precise instrument to forecast the Fintech index (Liu
et al., 2021). High frequency trading (HFT) algorithms are robust and
effective, this is shown by the fact that the whole prediction system,
which includes the deep learning block with RL framework corrections,
can boost trend forecast accuracy to roughly 85% (Rundo, 2019).
Additional insights on network connections may be gained by utilizing
the high-resolution information contained in high-frequency intraday
trading data sets. However, the asynchronicity, complicated dynamics,
and non-stationarity of such data sets make them extremely difcult to
model. Use of random forests, a cutting-edge machine learning approach
that provides high prediction accuracy without the need for costly
hyperparameter tweaking, to estimate nancial networks and overcome
these obstacles (Karpman et al., 2023).
Seven machine learning algorithms are compared in (Subasi et al.,
2021) study across four stock index datasets (NASDAQ, NYSE, NIKKEI,
and FTSE) with the goal of making investment risk mitigation easier.
Furthermore, Random Forest were found to produce superior outcomes.
Utilizing several types of random forests, including quantile random
forests and extreme random forests (Demirer et al., 2022), demonstrate
that risk aversion enhances the accuracy of realized volatility forecasts
outside of the sampling frame. Realized skewness and kurtosis, as well as
measures of jump intensity and leverage, have little to no effect on risk
aversion’s ability to anticipate future outcomes. Stock trend prediction
accuracy can be effectively improved by using the random forest model
and optimizing the various processes of stock research (Yin et al., 2023).
Random forest do the best job of predicting both the overalltrend and
the magnitude of future price changes (Akyildirim et al., 2022). Sador-
sky (2021) nds that, compared to logit models, random forest methods
are superior at predicting the direction of stock prices. For forecast ho-
rizons of 10 days or longer, random forests and tree-bagging models
have a prediction accuracy of above 80%. Accuracy ratings of 85–90%
are achieved by the tree bagging and random forest methods for a
20-day forecast horizon, whereas those of 55–60% are achieved by the
logit models. By applying the articial intelligence algorithm of
ensemble random forest methods, moreover, (Lin et al., 2020) was able
to get the probabilities of market reactions of start-up enterprises listed
on the GISA equity crowdfunding platform and anticipate the degree of
market reaction. Predicting actual data in market reactions using the
GISA platform for startups has an accuracy of 65%. Similarly, according
to (Luong and Dokuchaev, 2018), the random forest method is effective
at predicting the direction of realized volatility. Empirical results on the
S&P 200 indicate that the use of puried implied volatility and this
machine learning technique resulted in an improvement of the
pre-existing heterogeneous autoregressive model (HAR) framework.
2.1. Research gap
From the review of the existing literature, it can be observed that
many researches have been done on forecasting but there are some
studies related to forecasting stock prices using random forest but none
of the study focused on forecasting stock prices of Fintech Companies of
India using Random forest. Moreover, there is not enough studies that
used high-frequency data while framing random forest model for fore-
casting. The study is fully focused on a new dimension by framing
random forest model using the high-frequency data to forecast stock
prices of the companies of most emerging sector of India i.e., FinTech.
Hence, a study on this research gap is considered as a feasible one is
denitely going to contribute to existing studies.Moreover, studying
stock price forecasting in the context of ntech companies can provide
insights into how market dynamics are inuenced by innovation in the
nancial technology sector. Innovation Dynamics research may seek to
understand the interactions between technological innovation and
market behavior.
B.K. Meher et al.

Journal of Open Innovation: Technology, Market, and Complexity 10 (2024) 100180
3
3. Objectives of the study
•To formulate stock forecasting models for top three Fintech Com-
panies of India i.e., using Random Forest model with high-frequency
data.
•To determine the efcacy of the formulated models in forecasting the
future stock prices of ntech companies.
4. Research methodology
4.1. Research design
This research adopts a quantitative approach aimed at predicting
stock prices of Fintech companies in India. The chosen methodology
involves the application of a machine learning technique i.e. Random
Forest, implemented using Python. This method is suitable for fore-
casting nancial time series data as it accommodates non-linear re-
lationships and handles high-dimensional data effectively.
4.2. Data collection
The primary data source for this study consists of high-frequency
data i.e., historical one-minute daily open, high, low, close stock price
of top three Fintech companies listed on the Indian stock exchanges
namely Policy Bazar, One 97 Communications Paytm Ltd., and Niyogin
Ltd. Such kind of high-frequency data are proved as rich in quality
specically in forecasting stock prices due its precision and use in algo-
trading.High-frequency data in the context of the stock market refers to
data that is recorded and updated at very short time intervals, often with
sub-second or intra-second precision. This data can provide valuable
insights for stock market research, trading, and analysis.The data is
extracted from reputable nancial databases, such as www.money-
control.com and www.kotaksecurities.com, for a period from 1st
October, 2022–30 th September, 2023, ensuring the availability of a
sufcient historical dataset to train and validate the model. The data set
of each company has 97,760 data points. Hence, the study deals about
293,280 data points i.e., 3 companies @ 97,760 each.Innovation Dy-
namics research can involve studying new methods and tools for gath-
ering and interpreting data to drive innovation and the use of high-
frequency data and machine learning algorithms like Random Forest
in this research indicates an innovation in data collection and analysis
methods.
4.3. Data preprocessing
To prepare the data for model development, several preprocessing
steps are employed. These include data cleaning, feature selection,
handling missing values, and scaling. Stock price returns are calculated
and used as the target variable, while the lags of opening price, high
price and low price are used as independent variables.
4.4. Model development
Random Forest, a powerful ensemble machine learning algorithm, is
chosen as the forecasting model. It is known for its ability to handle
complex relationships and mitigate overtting. The model is imple-
mented using Python’s scikit-learn library. It is trained on historical
data, and hyperparameters are tuned through cross-validation to opti-
mize performance.
4.5. Softwares and applications used
For the formulation of Random Forest Model to forecast the stock
prices of Fintech Companies of India, PyCharm with Python Software
with version 3.8 (6) has been used. Moreover, various library packages
are installed namely, NumPy for numerical and mathematical
operations, Pandas for reading and cleaning the nancial data, Scikit-
Learn (sklearn) that includes Random Forest Regressor, Matplotlib and
Seaborn, openpyxl and xlrd for data visualization, High-frequency data
source, and pycharm. The selected version of python can smoothly
handle all the selected packages.
4.6. Evaluation techniques
To assess the model’s forecasting accuracy, several evaluation met-
rics are employed. These metrics include Mean Absolute Error (MAE),
Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and R-
squared (R2). Additionally, the out-of-sample forecasting accuracy is
assessed to validate the model’s generalization performance. The data is
split into training and testing sets, with a rolling window approach to
evaluate the model’s performance over time. 70% of the data i.e., 68432
are training data points and remaining 30% i.e., 29,328 data points are
used as testing sets.
By following the above approach, the study aims to develop an ac-
curate and robust forecasting model that can assist investors, analysts,
and policymakers in making informed decisions in the dynamic and
rapidly evolving Fintech sector of the Indian stock market.
5. Need of the study and managerial implications
A research study focused on forecasting stock prices of Fintech
companies in India using the Random Forest algorithm holds signicant
potential for societal benet. Accurate stock price predictions for these
companies are essential not only for investors but also for the broader
society. Firstly, such research aids investors in making informed de-
cisions. Stock market investments are integral to many individuals’
nancial planning, and the Fintech sector’s dynamism offers both op-
portunities and risks. Reliable forecasts can guide investors, helping
them allocate their resources effectively and mitigate potential losses.
This, in turn, promotes nancial literacy and stability among the
populace. Secondly, a successful prediction model for Fintech stocks can
stimulate investment in this sector. Fintech companies drive innovation,
nancial inclusion, and economic growth. If investors have condence
in their ability to forecast stock prices accurately, they are more likely to
invest in these companies, fostering innovation and job creation. The
success or failure of ntech companies can have a signicant economic
impact. Innovation Dynamics research often delves into the economic
consequences of innovation. Analyzing stock price forecasting in ntech
can contribute to understanding how innovation in this sector affects the
broader economy.Furthermore, a well-developed forecasting model can
serve as a valuable tool for regulators and policymakers. It enables them
to monitor market stability and respond proactively to emerging chal-
lenges, ultimately safeguarding the interests of both consumers and in-
vestors. This research on forecasting stock prices of Fintech companies
using Random Forest is a critical endeavor that benets society by
empowering investors, fostering innovation, and aiding regulatory
oversight in a sector poised for transformative growth.
6. Limitations of the study
•Limited Generalizability: The ndings of the paper may be limited in
their applicability to ntech companies in India. Stock price pre-
diction models developed for one market or time period may not
generalize well to different markets or time frames. Moreover, it may
not explore the robustness of the model to changing market condi-
tions, and whether it continues to perform well in different market
regimes.
•Data Quality and Availability: The accuracy and reliability of stock
price data, especially high-frequency data, can be a limitation. Data
may contain errors, gaps, or inconsistencies that can affect the results
and generalizability of the model. Moreover, such High-frequency
data often comes with a lag, which can affect the practicality of
B.K. Meher et al.

Journal of Open Innovation: Technology, Market, and Complexity 10 (2024) 100180
4
real-time trading strategies based on the model’s forecasts. The paper
may not address the implications of this lag.
•Model Parameter Tuning: Random Forest models have hyper-
parameters that need to be tuned for optimal performance. The paper
may not explore the sensitivity of the model’s performance to
different hyperparameter settings.
•Model Interpretability: Random Forest models are often considered
as "black-box" models, making it challenging to interpret the reasons
behind specic predictions. The paper may not delve into model
interpretability techniques or insights into the importance of specic
features.
•Market Dynamics and External Factors: The paper is not adequately
considered external factors such as economic conditions, regulatory
changes, or geopolitical events that can signicantly impact stock
prices, particularly in the volatile ntech sector.
7. Analysis and discussion
The study employs the Random Forest (RF) algorithm, originally
introduced by Breiman in 2001 as an enhanced version of the decision
tree, to predict stock prices within the ntech sector in India. At its core,
Random Forests consist of multiple decision trees. During training,
Random Forests construct numerous individual decision trees, and the
predictions from these trees are combined to make the nal prediction.
This aggregation is achieved by considering the mode of the classes for
classication tasks or the mean prediction for regression tasks. Two key
parameters in the RF model can signicantly impact its performance: the
number of trees (ntree) and the number of candidate variables randomly
selected at each split (ntry ). A recommended value for ntry is p
3, where p
represents the number of input variables (Dudek, 2015).
To forecast a time-series Yt, the study employs the autoregressive
random forest (AR-RF) model, denoted as AR-RF(p), where p signies
the number of autoregressive lags. In contrast to various other machine
learning models, Random Forests offer superior precision and excel in
handling large datasets with numerous variables, often extending into
the thousands. Furthermore, Random Forests possess the capability to
automatically balance datasets, especially when one class is less
frequent than others in the data.
7.1. Steps of applying random forest in predicting stock prices of ntech
companies
•In this study all the data sets of the ntech companies which are non-
stationary in nature are converted into stationary for the application
of random forest.
•Again, the data set of each company i.e., 97,760 data points, is
splitted into two parts, the rst part is the 70% contains 68,432 data
points and remaining 30% contains 29,328 data points.
•The rst part containing 68,432 data points is considered as training
set and the remaining part containing 29,328 data points is consid-
ered as testing set. The training set is used for framing the random
forest model and the testing set is where the formulated model is
applied for predicting.
•The strength and suitability of the random forest model can be
judged by values of Mean Squared Error (MSE), Root Mean Squared
Error (RMSE), Mean Absolute Error (MAE) and Coefcient of
Determination (R2)which can be calculated by using the following
formulae:-
Mean Squared Error Root Mean Squared Error
MSE =1
N∑N
i=1(yi−
y)2 RMSE =
MSE
√=
1
N∑N
i=1(yi−
y)2
√
Mean Absolute Error Coefcient of Determination
MAE =1
N∑N
i=1yi−
yR2=1−Σ(yi−
y)2
Σ(yi−y)2
Where 
y=predicted value of y and y=mean value of y
•After the application of the formulated model in the testing test, the outcomes of the
model in the form of predicted stock prices and the existing observed (actual) stock
prices are represented in graph to mark out the differences between the those two.
•As the testing set contains huge data i.e., 29,328 which may lack in clear depiction of
differences between the actual and predicted stock prices if presented through graphs.
Hence, a deviation graph has been used to show the deviations between the actual and
predicted stock prices.
7.2. Parameters for applying random forest
This research used “scikit-learn” Python packagefor formulating the
Random Forest model. There are various “hyperparameters” which
control the behavior of the algorithm and form the random forest model.
Research used default “hyperparameters” given in “scikit-learn” Python
package.
•Tree Count (n_estimators): Set to scikit-learn’s default value of 100,
which strikes a balance between model performance and computa-
tional economy.
•Trees’ Maximum Depth (max_depth): Left at ‘None’. This will let the
trees to grow until all of their leaves are pure or contain less samples
than min samples split. This helps in fully representing the
Complexity of the data, with further regularization coming from
other hyperparameters to reduce overtting.
•Minimum Samples for Split (min_samples_split): Set to 2, which is the
bare minimum needed to create a new node. As a result, the dataset
might be segmented nely, enhancing the model’s capacity to learn
from the training set.
•Minimum Samples (min_samples_leaf) at Leaf Nodes: Set to 1 allows
for the most detailed class denitions per leaf, especially helpful for
datasets with a complex decision boundary.
•Maximum Features (max_features): By default, "auto" is set in “scikit-
learn” Python package and the same is used, which chooses the
square root of the feature count. this parameter is essential in order
to diversify the individual trees and advance model generalization.
•Bootstrap Samples: By default, this feature is enabled in “scikit-
learn” Python package, this enables every tree to undergo training
using a bootstrapped sample of the data. By adding randomness to
the model, this lessens overtting and improves the ensemble’s ca-
pacity to generalize.
7.3. Brief about the selected ntech companies
7.3.1. Niyogin ntech Ltd
Niyogin Fintech Ltd. operates as a non-banking nance company,
specializing in providing loans, nancing, investments, and related
services to micro, small, and medium enterprises in India. The company
places a strong emphasis on superior execution, utilizing advanced
technology, innovative risk management, and establishing robust on-
ground connections. Niyogin’s primary objective is to offer small busi-
nesses an efcient and cost-effective support system through cutting-
edge technology and a dedicated network of partners. They aspire to
become the leading organization in India that caters to the needs of
small businesses, empowering their customers with a comprehensive
ecosystem of products, partnerships, technology, and exceptional
customer experiences.
One 97 Communications Ltd (Paytm).
According to RedSeer, One 97 Communications Ltd., better known as
Paytm, is India’s most advanced e-commerce platform. As of June 30,
2021, Paytm served 337 million clients and over 21.8 million registered
merchants with its vast suite of services, which included payment so-
lutions, e-commerce services, cloud services, and nancial services.
Paytm, a digital payment platform that was initially released in 2009
with a focus on mobile devices, ushered in a new era of cashless trans-
actions in India. Paytm Wallet rst appeared as a means of making
cellphone recharges and bill payments. Based on RedSeer’s analysis of
consumer volume, merchant count, transaction volume, and income,
Paytm had become India’s largest payments platform by March 31,
B.K. Meher et al.

Journal of Open Innovation: Technology, Market, and Complexity 10 (2024) 100180
5
2021. Paytm’s brand value of US$6.3 billion, according to the Kantar
BrandZ India 2020 Report, conrms its place as India’s most valuable
payments brand and makes it the go-to option for transactions across
multiple channels.
7.3.2. PB ntech Ltd. (policybazar)
Policybazaar’s parent business, PB Fintech Ltd., launched its agship
platform in 2008 to satisfy customers’ demands for improved insurance
education, selection, and disclosure. With the launch of Paisabazaar in
2014, the company set out to streamline the process of securing personal
loans and credit cards for individuals in India by putting an emphasis on
simplicity, speed, and openness. According to research from Frost &
Sullivan, in terms of disbursals in Fiscal 2021, Paisabazaar held a
dominant 53.7% market share in India’s digital consumer credit
marketplace. In addition, Policybazaar overtook all other online insur-
ance distributors in Fiscal 2020 to become the largest digital insurance
marketplace, with a market share of 93.4% based on the number of
policies sold. Also, when looking at the total number of policies sold
online in India, both insurance rms and distributors, Policybazaar
accounted for 65.3%. Following this section is a visual representation of
the real stock values of a few India-based ntech companies.
Graphical Representation of actual Stock Prices of Selected
Fintech Companies.
.
Source: Authors’ Construction of Graphs using EVIEWS.
The graphs of the stock prices of all the selected ntechs seem non-
stationary, hence, it is necessary to convert the non-stationary data
into stationary data for the application of random forest for which log
returns have been computed and plotted using line graphs which are
shown in the succeeding section.
Graphical Representation of Log Returns of Selected Fintech
Companies.
.
20
30
40
50
60
70
80
90
100
October January April July
32022202
Stock Prices of Niyogin Ltd.
400
500
600
700
800
900
1,000
October January April July
32022202
Stock Prices of On e 97 Communica�ons Paytm Ltd.
300
400
500
600
700
800
900
October January April July
32022202
Stock Prices of Policy Bazar Ltd.
Table 1
Values of Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE) and Coefcient of Determination (R
2)using Python.
Statistics Niyogin Fintech Ltd. One 97 Communications Ltd (Paytm) PB Fintech Ltd. (Policybazar)
Mean Squared Error (MSE) 0.00598 0.1502 0.1486
Root Mean Squared Error (RMSE) 0.07736 0.3875 0.3855
Mean Absolute Error (MAE) 0.01148 0.2337 0.2336
Coefcient of Determination (R2)0.99998 0.99999 0.99999
Source: Authors’ Computation using Python
B.K. Meher et al.

Journal of Open Innovation: Technology, Market, and Complexity 10 (2024) 100180
6
Source: Authors’ Construction of Graphs using EVIEWS.
The line graphs of log returns seem stationary in nature. Moreover,
the stationary of data should be examined through statistical test.
Hence, the stationarity of log returns series of the above ntech com-
panies have been examined with the help of a unit root test named
Augmented Dickey Fuller Test with the inclusion of test equation as
Intercept, Trend and Intercept and None and found stationary and now
prepared to use it for random forest model.
Table 1 depicts the calculated Mean Squared Error (MSE), Root Mean
Squared Error (RMSE), Mean Absolute Error (MAE) and Coefcient of
Determination (R2)of all the selected ntechs. MSE measures the
average squared difference between the predicted values and the actual
values. A lower MSE indicates that the model’s predictions are closer to
the actual values. The MSE of Niyogin Fintech Ltd., One 97 Communi-
cations Ltd (Paytm) and PB Fintech Ltd. (Policybazar) is 0.00598,
0.1502 and 0.1486 respectively which suggests that, on average, the
squared difference between the predicted and actual stock prices is quite
small. This is a positive sign for the model’s accuracy.
Table 1RMSE is the square root of the MSE. It provides a measure of
the average magnitude of errors in the same units as the predicted
variable. An RMSE of 0.07736, 0.3875 and 0.3855 of the three ntechs
respectively indicates that, on average, the model’s predictions are
approximately 0.07736, 0.3875 and 0.3855 units away from the actual
stock prices of the three ntechs respectively. In a suitable model a
lower RMSE is always desirable.
Similarly, the MAE measures the average absolute difference be-
tween the predicted and actual values. Like MSE and RMSE, a lower
MAE indicates better model accuracy and this is another indicator of
good model performance.
Moreover, the R
2
of all selected ntech are very close to 1, suggesting
that the models explains almost all of the variance in the stock prices.
This is an exceptionally high R-squared value and indicates an excellent
t of the model to the data. In summary, based on the provided statis-
tical results, it appears that the Random Forest model used for fore-
casting stock prices of all Fintech companies are performing
exceptionally well. The low MSE, RMSE, and MAE values indicate that
the model’s predictions are very close to the actual values. Additionally,
the high R-squared value (close to 1) suggests that the model is an
excellent t for the data.
Again, by using the formulated models the predicted stock prices are
calculated and shown in the line graphs along with the actual stock
prices for comparison.
Graphical Representation of Actual and Predicted Stock Prices
of Selected Fintech Companies on testing data using the formulated
Random Forest Model.
-.15
-.10
-.05
.00
.05
.10
October January April July
32022202
Log Returns of Niy ogin Fintech L td.
-.06
-.04
-.02
.00
.02
.04
.06
.08
October January April July
32022202
Log Retuns of One 97 Communica�ons Ltd (Paytm)
-.03
-.02
-.01
.00
.01
.02
.03
.04
.05
.06
October January April July
32022202
log Returns of P B Fintech Ltd. (Policybazar)
B.K. Meher et al.

Journal of Open Innovation: Technology, Market, and Complexity 10 (2024) 100180
7
B.K. Meher et al.

Journal of Open Innovation: Technology, Market, and Complexity 10 (2024) 100180
8
.
Source: Authors’ Construction of Graphs using Python.
Since the above graphs consist nearly 29,328 observations used for
prediction, the trend line of actual and predicted stock prices of the
companies cannot be precisely seen.
Hence, for simplicity and clear understanding, deviations plot (the
difference between actual and predicted stock prices) are generated that
could depict a single trend line. If the model’s predictions are perfect,
the difference will be zero, and no spike can be seen on that particular
date and time. Otherwise, the line will oscillate around zero, showing
the upward spike where predictions are high and downward spike where
predictions are low. The succeeding section shows the deviations plot of
all the three selected companies.
Graphical Representation of Deviations plot of Selected Fintech
Companies.
.
Source: Authors’ Construction of Graphs using Python.
From the deviation graphs it can be observed that there are very few
and small spikes in the Policy Bazar graph representing much accuracy
in the prediction and a very few i.e., around 5–6 spikes that oscillated
between –1.25 to +4 against the average price 576.70 in last year.
Again, some spikes can be seen in the Niyogin Ltd. graph with a
maximum oscillation of –3 to +2 where the average price of Niyogin
Ltd. in last one year is about 48, but still it can be considered as accurate
forecasting as most of the spikes in particular timestamp are near to zero.
Lastly, in Paytm Graph, too many spikes can be seen but the most of the
spikes are oscillating between –2.5 to +2.5 and very few spikes between
–12.5 to +5 where the average price in last one year is about 688 and
again it can be considered as optimally accurate forecasting.
8. Conclusion
From observation of above analysis and discussion it is clear that the
results obtained from this study provide valuable insights into the ac-
curacy and performance of the predictive model. The statistical metrics
presented in Table 1, including Mean Squared Error (MSE), Root Mean
Squared Error (RMSE), Mean Absolute Error (MAE), and Coefcient of
Determination (R2), serve as crucial indicators of the model’s effec-
tiveness. It is evident that the Random Forest model has demonstrated
exceptional predictive capabilities for all the selected ntech companies.
First, the low MSE values for Niyogin Fintech Ltd., One 97 Communi-
cations Ltd (Paytm), and PB Fintech Ltd. (Policybazar) indicate that the
model’s predictions are consistently close to the actual stock prices. This
reects the accuracy of the model in capturing the underlying patterns in
the stock price movements. The RMSE values further conrm the
model’s accuracy, with values of 0.07736, 0.3875, and 0.3855 for the
three ntechs, respectively. These results indicate that, on average, the
model’s predictions deviate by a relatively small margin from the actual
stock prices. Lower RMSE values are generally preferred in forecasting
models, and in this case, they signify the model’s reliability in making
precise predictions. Additionally, the MAE values for all selected n-
techs are low, reafrming the model’s ability to make accurate pre-
dictions with minimal absolute errors. This is a crucial aspect of model
performance, as lower MAE values imply better accuracy in predicting
stock prices. Perhaps the most compelling evidence of the model’s
prociency is the high R^2 values, which are very close to 1 for all n-
tech companies. An R2 value near 1 indicates that the model explains
almost all of the variance in the stock prices, demonstrating an excep-
tional t of the model to the data. This is a remarkable achievement, as it
signies that the Random Forest model effectively captures the under-
lying factors inuencing the stock prices of these ntech companies.
Furthermore, the deviation graphs provided in the analysis offer visual
B.K. Meher et al.

Journal of Open Innovation: Technology, Market, and Complexity 10 (2024) 100180
9
conrmation of the model’s accuracy. The small and infrequent spikes in
the Policy Bazar graph, the limited oscillations in the Niyogin Ltd. graph,
and the relatively manageable spikes in the Paytm graph all contribute
to the overall assessment of accurate forecasting. Moreover, the ndings
of this research paper strongly support the effectiveness of the Random
Forest model in forecasting stock prices of ntech companies in India
using high-frequency data. The combination of low MSE, RMSE, and
MAE values, along with the high R2 values and visually conrmed ac-
curacy, underscores the model’s exceptional performance. These results
have signicant implications for investors, nancial analysts, and
decision-makers, as they can rely on this model to make informed de-
cisions in the dynamic ntech sector. This study contributes to the
growing body of knowledge in nancial forecasting and reinforces the
value of machine learning techniques in stock price prediction.But there
are also certain limitations of the ndings of this study, like the results
and ndings are based on the lagged open, high and low stock prices
only and does not consider any macro-economic factors that might affect
the stock prices of ntech companies in India.Moreover, in last few years
the nancial inclusion has been signicantly increased. Many people of
India have started using ntech services especially the digital banking
transactions after demonetization and during COVID-19, which ulti-
mately shoot up the demand for services of ntech companies that leads
to increment of retail investment on such companies, but such effect are
also not considered as regressor.Hence, the future researchers may
conduct some study which could overcome such limitations too.
Furthermore, this study could also act as a guide for future researchers to
employ varied nature of random forest using high-frequency data on
unexplored sectors and areas of global stock markets to forecast the
stock prices or values of market indices.
Author contributions
All authors contributed equally to this research work. All authors
discussed the results and contributed to the nal manuscript. All authors
have read and agreed to the published version of the manuscript.
Ethical statement/approval
No applicable because the study does not include research involving
animal or human subjects.
Funding
This research received no external funding.
CRediT authorship contribution statement
Conceptualization: Bharat Kumar Meher. Data curation: Bharat
Kumar Meher, Manohar Singh, Abhishek Anand, Ramona Birau. Formal
analysis: Bharat Kumar Meher, Manohar Singh, Ramona Birau, Investi-
gation: Bharat Kumar Meher, Manohar Singh, Ramona Birau. Review of
Literature and Research Gap: Abhishek Anand, Bharat Kumar Meher.
Methodology: Bharat Kumar Meher, Manohar Singh, Abhishek Anand.
Project administration: Bharat Kumar Meher, Ramona Birau. Resources:
Ramona Birau. Software: Manohar Singh, Bharat Kumar Meher. Vali-
dation: Bharat Kumar Meher, Abhishek Anand. Visualization: Manohar
Singh, Ramona Birau. Writing – original draft: Bharat Kumar Meher,
Abhishek Anand. Writing – reviewing & editing: Ramona Birau, Bharat
Kumar Meher, Abhishek Anand, Manohar Singh.
Declaration of Competing Interest
The authors declare that they have no known competing nancial
interests or personal relationships that could have appeared to inuence
the work reported in this paper.
Acknowledgement
This research did not receive any specic grant from funding
agencies in the public, commercial, or not-for-prot sectors.
References
Akyildirim, E., Bariviera, A.F., Nguyen, D.K., Sensoy, A., 2022. Forecasting high-
frequency stock returns: a comparison of alternative methods. Ann. Oper. Res. 313
(2), 639–690. https://doi.org/10.1007/s10479-021-04464-8.
Akyildirim, E., Nguyen, D.K., Sensoy, A., ˇ
Siki´
c, M., 2023. Forecasting high-frequency
excess stock returns via data analytics and machine learning. Eur. Financ. Manag. 29
(1), 22–75. https://doi.org/10.1111/eufm.12345.
Alkhatib, K., Khazaleh, H., Alkhazaleh, H.A., Alsoud, A.R., Abualigah, L., 2022. A new
stock price forecasting method using active deep learning approach. J. Open Innov.:
Technol., Mark., Complex., 8(2), Artic. 2. https://doi.org/10.3390/joitmc8020096.
Alt, R., Beck, R., Smits, M.T., 2018. FinTech and the transformation of the nancial
industry. Electron. Mark. 28 (3), 235–243. https://doi.org/10.1007/s12525-018-
0310-9.
Arora, S., Madan, P., 2023. Conceptual framework depicting the drivers for the ntech
growth: an outlook for India. In: Grima, S., Sood, K., ¨
Ozen, E. (Eds.), Contemporary
Studies of Risks in Emerging Technology, Part A. Emerald Publishing Limited,
pp. 197–220. https://doi.org/10.1108/978-1-80455-562-020231014.
Asad, M., 2015. Optimized Stock market prediction using ensemble learning. 9th Int.
Conf. Appl. Inf. Commun. Technol. (AICT) 2015, 263–268. https://doi.org/10.1109/
ICAICT.2015.7338559.
Bhatia, A., Chandani, A., Atiq, R., Mehta, M., Divekar, R., 2021. Articial intelligence in
nancial services: a qualitative research to discover robo-advisory services. Qual.
Res. Financ. Mark. 13 (5), 632–654. https://doi.org/10.1108/QRFM-10-2020-0199.
Bhatnagar, M., ¨
Ozen, E., Taneja, S., Grima, S., Rupeika-Apoga, R., 2022. The dynamic
connectedness between risk and return in the ntech market of India: evidence using
the GARCH-M approach. Article 11 Risks 10 (11). https://doi.org/10.3390/
risks10110209.
Breidbach, C.F., Keating, B.W., Lim, C., 2020. Fintech: research directions to explore the
digital transformation of nancial service systems. J. Serv. Theory Pract. 30 (1),
79–102. https://doi.org/10.1108/JSTP-08-2018-0185.
Brown, K.C., Wiles, K.W., 2020. The growing blessing of unicorns: the changing nature of
the market for privately funded companies. J. Appl. Corp. Financ. 32 (3), 52–72.
https://doi.org/10.1111/jacf.12418.
Chatterjee, A., Bhowmick, H., Sen, J., 2021. Stock price prediction using time series,
econometric, machine learning, and deep learning models. IEEE Mysore Sub Sect.
Int. Conf. (MysuruCon) 2021, 289–296. https://doi.org/10.1109/
MysuruCon52639.2021.9641610.
Demirer, R., Gkillas, K., Gupta, R., Pierdzioch, C., 2022. Risk aversion and the
predictability of crude oil market volatility: a forecasting experiment with random
forests. J. Oper. Res. Soc. 73 (8), 1755–1767. https://doi.org/10.1080/
01605682.2021.1936668.
Dudek, G., 2015. Short-term load forecasting using random forests. In: Filev, D., Jab
\lkowski, J., Kacprzyk, J., Krawczak, M., Popchev, I., Rutkowski, L., Sgurev, V.,
Sotirova, E., Szynkarczyk, P., Zadrozny, S. (Eds.), Intelligent Systems’2014. Springer
International Publishing, pp. 821–828.
Feyen, E., Frost, J., Gambacorta, L., Natarajan, H., & Saal, M. (2021). Fintech and the
digital transformation of nancial services: Implications for market structure and
public policy. https://www.bis.org/publ/bppdf/bispap117.htm.
Ganjoo, D., Mukherjee, S., Mukhopadhyay, S., 2023. Razorpay: providing payment
convenience to disruptors. Indian Inst. Manag. Ahmedabad 1–19. https://doi.org/
10.1108/CASE.IIMA.2023.000009.
Harris, J.L., 2021. Bridging the gap between ‘Fin’ and ‘Tech’: the role of accelerator
networks in emerging FinTech entrepreneurial ecosystems. Geoforum 122, 174–182.
https://doi.org/10.1016/j.geoforum.2021.04.010.
Jakhiya, M., Mittal Bishnoi, M., Purohit, H., 2020. Emergence and growth of mobile
money in modern india: a study on the effect of mobile money. Adv. Sci. Eng.
Technol. Int. Conf. (ASET) 2020, 1–10. https://doi.org/10.1109/
ASET48392.2020.9118375.
Karpman, K., Basu, S., Easley, D., Kim, S., 2023. Learning nancial networks with high-
frequency trade data. Data Sci. Sci. 2 (1), 2166624 https://doi.org/10.1080/
26941899.2023.2166624.
Khando, K., Islam, M.S., Gao, S., 2023. The emerging technologies of digital payments
and associated challenges: a systematic literature review. Future Internet 15 (1).
https://doi.org/10.3390/15010021.
Landis, W., Cha, S., 2020. Towards high performance stock market prediction methods.
IEEE Cloud Summit 2020, 156–160. https://doi.org/10.1109/
IEEECloudSummit48914.2020.00030.
Lin, C.-S., Lin, C.-Y., Reynolds, S., 2020. Applying the random forest model to forecast
the market reaction of start-up rms: case study of GISA equity crowdfunding
platform in Taiwan (Scopus). WSEAS Trans. Bus. Econ. 17, 241–259. https://doi.
org/10.37394/23207.2020.17.26.
Liu, C., Fan, Y., Zhu, X., 2021. Fintech index prediction based on RF-GA-DNN algorithm.
Wirel. Commun. Mob. Comput. 2021, e3950981 https://doi.org/10.1155/2021/
3950981.
Luong, C., Dokuchaev, N., 2018. Forecasting of realised volatility with the random
forests algorithm. J. Risk Financ. Manag. 11 (4), 4 https://doi.org/10.3390/
jrfm11040061. Article 4.
B.K. Meher et al.

Journal of Open Innovation: Technology, Market, and Complexity 10 (2024) 100180
10
Mention, A.-L., 2019. The future of ntech. Res. Technol. Manag. 62 (4), 59–63. https://
doi.org/10.1080/08956308.2019.1613123.
Migozzi, J., Urban, M., W´
ojcik, D., 2023. You should do what India does”: FinTech
ecosystems in India reshaping the geography of nance. Geoforum, 103720. https://
doi.org/10.1016/j.geoforum.2023.103720.
Mohanta, B., Nanda, P., Patnaik, S., 2020. Management of V.U.C.A. (Volatility,
Uncertainty, Complexity and Ambiguity) using machine learning techniques in
industry 4.0 paradigm. In: Patnaik, S. (Ed.), Big Data & Cyber Physical Systems (pp,
New Paradigm of Industry 4.0: Internet of Things. Springer International Publishing,
pp. 1–24. https://doi.org/10.1007/978-3-030-25778-1_1.
Moid, S., & Shankar, N. (2022). Creating Value Proposition for Rural Banking Customers
in Emerging Markets: Adoption of Mobile Banking Technology Induced by
Disruptive Events in India. In A. Thrassou, D. Vrontis, L. Efthymiou, Y. Weber, S. M.
R. Shams, & E. Tsoukatos (Eds.), Business Advancement through Technology Volume
I: Markets and Marketing in Transition (pp. 47–72). Springer International
Publishing. https://doi.org/10.1007/978–3-031–07769-2_3.
Nenavath, S., 2022. Impact of ntech and green nance on environmental quality
protection in India: By applying the semi-parametric difference-in-differences
(SDID). In: Renewable Energy, 193, pp. 913–919. https://doi.org/10.1016/j.
renene.2022.05.020.
Palmi´
e, M., Wincent, J., Parida, V., Caglar, U., 2020. The evolution of the nancial
technology ecosystem: An introduction and agenda for future research on disruptive
innovations in ecosystems. Technol. Forecast. Soc. Change 151, 119779. https://doi.
org/10.1016/j.techfore.2019.119779.
Pazarbasioglu, C., Mora, A.G., Uttamchandani, M., Natarajan, H., Feyen, E., & Saal, M.
(2020). DIGITAL FINANCIAL SERVICES.
Priyamvada, Wadhvani, R., 2017. Review on various models for time series forecasting.
2017 Int. Conf. Invent. Comput. Inform. (ICICI) 405–410. https://doi.org/10.1109/
ICICI.2017.8365383.
Rundo, F., 2019. Deep LSTM with reinforcement learning layer for nancial trend
prediction in FX high frequency trading systems. Appl. Sci. 9 (20), 20. https://doi.
org/10.3390/app9204460. Article 20.
Sadorsky, P., 2021. A random forests approach to predicting clean energy stock prices.
J. Risk Financ. Manag. 14 (2), 2 https://doi.org/10.3390/jrfm14020048. Article 2.
Saura, J.R., Reyes-Men´
endez, A., deMatos, N., Correia, M.B., 2021. Identifying startups
business opportunities from UGC on twitter chatting: an exploratory analysis.
J. Theor. Appl. Electron. Commer. Res. 16 (6), 1929–1944. https://doi.org/10.3390/
jtaer16060108.
Scardovi, C., 2017. Digital Transformation in Financial Services. Springer,.
Selvamuthu, D., Kumar, V., Mishra, A., 2019. Indian stock market prediction using
articial neural networks on tick data. Financ. Innov. 5 (1), 16 https://doi.org/
10.1186/s40854-019-0131-7.
Subasi, A., Amir, F., Bagedo, K., Shams, A., Sarirete, A., 2021. Stock market prediction
using machine learning. Procedia Comput. Sci. 194, 173–179. https://doi.org/
10.1016/j.procs.2021.10.071.
Surana, K., Singh, A., Sagar, A.D., 2020. Strengthening science, technology, and
innovation-based incubators to help achieve sustainable development goals: lessons
from India. Technol. Forecast. Soc. Change 157, 120057. https://doi.org/10.1016/j.
techfore.2020.120057.
Taujanskait˙
e, K., Kuizinait ˙
e, J., 2022. Development of FinTech business in Lithuania:
driving factors and future scenarios. Bus. Manag. Econ. Eng. 20 (1), 1 https://doi.
org/10.3846/zenodo.2022.16738.
Thakur, M., Kumar, D., 2018. A hybrid nancial trading support system using multi-
category classiers and random forest. Appl. Soft Comput. 67, 337–349. https://doi.
org/10.1016/j.asoc.2018.03.006.
Valencia, F., G´
omez-Espinosa, A., Vald´
es-Aguirre, B., 2019. Price movement prediction
of cryptocurrencies using sentiment analysis and machine learning. Entropy 21 (6),
6. https://doi.org/10.3390/e21060589.
Yin, L., Li, B., Li, P., Zhang, R., 2023. Research on stock trend prediction method based
on optimized random forest. CAAI Trans. Intell. Technol. 8 (1), 274–284. https://
doi.org/10.1049/cit2.12067.
Zhang-Zhang, Y., Rohlfer, S., Rajasekera, J., 2020. An eco-systematic view of cross-sector
ntech: the case of Alibaba and Tencent. Sustainability 12 (21). https://doi.org/
10.3390/su12218907. Article 21.
B.K. Meher et al.