Abstract
Purpose
The purpose of this paper is to propose a novel model, to forecast demand for a third-party service by using the Grey Systems Theory (GST) and Markov Chains, where its forecast error performance is evaluated through mean percentage error, mean absolute percentage error, where it exceed other models' performance accuracy such as autoregressive integrated moving average.
Design/methodology/approach
The model performs data characterization to qualify the GM (1,1) model and then applies a Markov Chain transition probability matrix and the GM (1,1) to forecast a time series with high degree of vagueness and imprecision and by providing a forecast kernel range .
Findings
The MCGM (1,1) model integrates the GST GM (1,1) and Markov Chains in a novel hybrid model, that reduces the mathematical calculation complexity while provides practical forecast performance that exceed or it is equally good as other traditional methods.
Research limitations/implications
The model outperforms other non-stationary models but does not incorporate multiple variables and requires additional mathematical treatment or combined methods, where its data is stationary, seasonal or negative.
Practical implications
This model can provide an accurate forecast projection of supply chain demand, for instance the space required in a third-party logistics services provider in Tijuana Mexico, it can be used to forecast complex supply chain systems with minimum, incomplete or poor data, to solve several practical application problems to forecast demand and resources.
Social implications
The novel MCGM (1,1) hybrid forecasting model combines multiple predictive approaches, allowing for greater accuracy and adaptability. Its implementation enhances decision-making in key sectors such as health, energy and manufacturing, optimizing resources and reducing costs. This drives economic growth, increases sustainability and improves the quality of life in society.
Originality/value
There are no MCGM (1,1) works applied in supply chain and current works have not established the model characterization criteria. The result of this investigation represents a novel proposal to solve uncertain models with poor information and small amounts of data (>4 records), with higher forecast accuracy.