Predicting Indian electricity exchange‐traded market prices: SARIMA and MLP approach

• Published date: 01 December 2023
• Author: Sonal Gupta,Deepankar Chakrabarty,Rupesh Kumar
• Date: 01 December 2023
• DOI: http://doi.org/10.1111/opec.12287

OPEC Energy Rev. 2023;47:271–286.

|

271wileyonlinelibrary.com/journal/opec

1

|

INTRODUCTION

Electricity use is indispensable for economic growth and almost all the areas of development, including social, economic

and, even individual livelihoods. The emergent electricity market like India has undergone various regulatory changes

over the past two decades. With an inception of the Electricity Act 2003, the power market has been transformed into a

competitive, efficient, and transparent mechanism (Ahmad & Alam,2019). An introduction of smart grids, electric ve-

hicles and a requirement to integrate renewable energy into the grid (to achieve an ambitious goal of renewable energy

of 50% in India's total electricity mix by 2030), have led to a need to forecast electricity price for energy systems planning

and its operations. Despite these policy initiatives and amendments in the market structure, this niche market still experi-

ences various issues such as a huge demand– supply gap, instability in the electrical grid, lesser volume traded in evolved

markets and lack of efficient risk management practices (Roy & Basu,2020). When compared with the volumes traded

in the European exchange- traded market, which range from 23% to as high as 88% in the year 2015 (Gupta et al.,2020;

Sharma et al.,2023), it represents a meager 3– 4% of the total power generation as of 2019– 2020 in India. This low trading

volume leaves room for a greater scope of research in the Indian ST electricity market. The above- mentioned problems

can be solved with the application of robust price forecasting tools, making it a dependable trading platform. Electricity

in itself is a complex commodity. Due to its distinctive characteristics such as inelastic demand, restrictive transportation

networks, kinked supply curve, weather dependence and non- storability (Robert & Mount,1998), there is higher insta-

bility, sharp cost spikes and mean- returning behaviours (Gupta et al.,2020), exposing the various electric utilities such

as producers, traders and others to many kinds of risks. According to Central Electricity Regulatory Commission (2021),

DOI: 10.1111/opec.12287

ORIGINAL ARTICLE

Predicting Indian electricity exchange- traded market

prices: SARIMA and MLP approach

SonalGupta1,2

|

DeepankarChakrabarty3

|

RupeshKumar4

© 2023 Organization of the Petroleum Exporting Countries.

1School of Business, UPES, Dehradun,

India

2School of Engineering, Institute

for Energy Systems, University of

Edinburgh, Edinburgh, UK

3Jaipuria Institute of Management,

Noida, India

4Jindal Global Business School (JGBS),

O.P. Jindal Global University, Sonipat,

India

Correspondence

Rupesh Kumar, Jindal Global Business

School (JGBS), O.P. Jindal Global

University, Sonipat, Haryana, India.

Email: scholar.rupesh@gmail.com

Abstract

This research investigates the short- term (ST) forecasting performance of the

daily prices of the Indian exchange- traded day- ahead (DAM) market, divided into

13 bid areas, each consisting of states with varied fundamentals. Forecasts are

built employing SARIMA (seasonal autoregressive integrated moving average)

and MLP (multilayer perceptron) methods. Moreover, the robustness and perfor-

mance of the model is compared using the lowest error and the Diebold– Mariano

(DM) test statistic values. The results indicates that the SARIMA model has high

prediction accuracy with error values ranging from 1% to 5% with Southern re-

gion having the highest error of 4.53% and Northern having the least error of

1.27%. However, validation by the DM test suggests no statistical significant dif-

ference between the two models. The power generators, distribution companies,

traders, policymakers, strategists and managers could use the findings for effec-

tive power management through proper planning.

272

|

GUPTA et al.

volatility in Indian Energy Exchange prices have ranged from as high as more than 20% in 2008– 2009 to as low as around

7% in 2013– 2014, which have caused huge losses to the power market stakeholders. According to Patil(2017), Discoms

countrywide have experienced losses equivalent to 4% of India's GDP (gross domestic product) and around Rs. 68,000

crore ($10 billion) each year. Hence, accurate electricity price modelling has become a challenging task. Although, nu-

merous studies have been conducted with an application of machine learning (ML) and statistical methods, to find a con-

crete and robust model, yet these studies' conclusions are contradictory. In some cases, innovative statistical techniques

are matched with simple ML methods (Marcjasz et al.,2019) and draw that statistical methods are better. At the same

time, studies suggesting novel ML techniques compare them with simple statistical approaches (Chen et al.,2019; Luo &

Weng,2019) stating ML models to be more precise. Hence, this research aims to evaluate the ST electricity price forecast

(EPF) (seven days) performance of evolved time- series models on one hand and the latest machine learning technique

on the other. Based on various calibrations, the models are evaluated by comparing the mean absolute error (MAE) and

mean absolute percentage error (MAPE) values. For analysis of every 7 days, we have also performed a significance test

(Diebold– Mariano (DM) test in our case) to authenticate the performance of every model applied. We have applied the

models on the Indian electricity exchange- traded market prices of 13 bid areas (shown in Table1) for ten years. Each bid

area consists of several states of India which has their distinctive demand– supply characteristics due to which they are

exposed to huge volatility.

Data were collected based on their availability. The presence of structural breaks is checked by employing the

Augmented Dickey– Fuller (ADF) test with a break using both additive and innovative outliers. Moreover, a seasonality

unit root (UR) test has been applied to the data series to check the presence of UR consistent with seasonal frequencies.

The outcome suggests that there is no statistical significant difference between the two models with MAPE values of the

SARIMA model being lower than that of the MLP model in all the bid areas.

The remainder of the paper is divided into four sections. Section2 entails a literature review providing a compre-

hensive view of the latest research in EPF; Section3 focuses on the data series used and the implemented research

TABLE IEX bid areas (Gupta et al.,2020).

S.No.Bid areaRegionsStates

1E1EasternWest Bengal, Sikkim, Bihar,

Jharkhand

2E2 EasternOrissa

3N1NorthernJammu and Kashmir, Himachal

Pradesh, Chandigarh,

Haryana

4N2NorthernUttar Pradesh, Uttaranchal,

Rajasthan, Delhi

5N3 NorthernPunjab

6A1NortheasternTripura, Manipur, Mizoram,

Nagaland

7A2NortheasternAssam, Arunachal Pradesh,

Meghalaya

8W1WesternMadhya Pradesh

9W2WesternMaharashtra, Gujarat, Daman

and Diu, Dadar and Nagar

Haveli, North Goa

10W3 WesternChhattisgarh

11S1SouthernAndhra Pradesh, Telangana,

Karnataka, Pondicherry

(Yanam), South Goa

12S2SouthernTamil Nadu, Pondicherry

(Puducherry), Pondicherry

(Karaikal), Pondicherry

(Mahe)

13S3 SouthernKerala