Impact of the Price of Petroleum on the Price of Food Items in Sri Lanka

The rising cost of food presents a significant challenge for households in developing countries, prompting experts to explore effective policies to mitigate price volatility. This concern has recently gained attention in Sri Lanka, sparking discussions on the current and future trajectory of food commodity prices. The increase in agricultural commodity prices is partially attributed to the rising costs of petroleum. This study aims to understand the relationship between coconut prices and fluctuations in the price of Lanka Petrol (LP92), as well as the connection between coconut prices and changes in the price of Lanka Auto Diesel (LAD). It also investigates variations in Samba prices concerning LP92 price changes, given their significance as highly consumed food products and widely used petroleum commodities. Utilizing data from January 2008 to November 2022 sourced from the official website of the Central Bank of Sri Lanka, the study employs Vector Auto Regressive (VAR) models. Model selection is determined by criteria such as Akaike Information Criterion (AIC), Schwarz-Bayesian Information Criterion (SC), and Hannan-Quinn (HQ). The study finds that VAR (2) is optimal for identifying Samba price variations with LP92, VAR (3) for Coconut price variations with LAD, and VAR (3) for Coconut price variations with LP92. The dataset is split into an 80% training set and a 20% test set for model evaluation, utilizing metrics like Root Mean Squared Error (RMSE), Mean Absolute Percentage Error (MAPE), and Mean Absolute Error (MAE). Further investigations into the impact of fuel costs on Samba and Coconut prices, using Granger causality and cointegration tests with LAD and LP92 price data, yield no conclusive evidence of causation. The study proceeds to remodel the original datasets using machine learning techniques, with the models demonstrating superior performance compared to traditional VAR models. Based on all the models, the models created through machine learning techniques showed superior performances showing 27%, 30%, 19% MAPE values and 11.12, 14.7, 16.6 MAE values for respective three FFNN models. The findings of this study would be more beneficial for policymakers, researchers, sellers as well as buyers.