Boost Time Series Forecast Accuracy with the Grey‑Markov Hybrid Model

Grey Model GM(1,1) Overview

Grey prediction models reduce randomness in original data by cumulative generation, revealing underlying patterns; GM(1,1) is the most commonly used first‑order single‑variable grey model.

The modelling steps are:

Generate sequence : define the original data series.

Accumulate sequence : obtain a new accumulated series.

Establish grey differential equation with unknown parameters.

Estimate parameters using least‑squares method.

Solve the differential equation to get the predicted series.

Restore forecast values by reversing the accumulation.

Markov Chain Overview

Markov chains describe system state transitions where the next state depends only on the current state, characterized by a state space and a transition probability matrix.

Key definitions:

State space : the set of all possible system states.

Transition probability matrix : matrix whose element P(i,j) denotes the probability of moving from state i to state j.

Grey‑Markov Model

The hybrid model combines the strengths of grey models and Markov chains: first use the grey model for an initial forecast, then correct the forecast with a Markov chain to improve accuracy.

Modeling steps:

Grey model prediction : apply the GM(1,1) model to the time series to obtain initial forecasts.

State division : partition the forecasted and actual values into several states based on historical data.

State transition probability calculation : compute the transition probability matrix from historical data.

Correct forecast values : adjust the grey model forecasts according to the state of each forecast and the transition probabilities.

Case study: forecasting the Consumer Price Index (CPI) for September–November 2011 using data from January 2010 to August 2011.

Data (month, CPI): 2010.01 1.5, 2010.02 2.7, 2010.03 2.4, 2010.04 2.8, 2010.05 3.1, 2010.06 2.9, 2010.07 3.3, 2010.08 3.5, 2010.09 3.6, 2010.10 4.4, 2010.11 5.1, 2010.12 4.6, 2011.01 4.9, 2011.02 4.9, 2011.03 5.4, 2011.04 5.2, 2011.05 5.3, 2011.06 5.5, 2011.07 6.4, 2011.08 6.2.

Step 1 – Grey model prediction: generate and accumulate the sequence, estimate parameters, forecast, and restore values.

Step 2 – State division: divide the CPI residual series into five states based on historical averages.

Step 3 – Transition probability calculation: derive the transition matrix from the state sequence.

Step 4 – Forecast correction: use the transition matrix to adjust the grey model forecasts. For example, if a grey forecast residual falls into state 4, the corrected forecast considers probabilities of moving to states 1‑5.

The GM(1,1) model alone yields a MAPE of 14.60 %; after Markov correction, the hybrid model’s MAPE drops to 7.32 %, clearly demonstrating the advantage of the Grey‑Markov approach for complex time‑series prediction.

References: [1] Wang Panyu. Application of Grey‑Markov Model in CPI Forecasting. Times Finance, 2012, (15):213‑214. [2] Lu Hongbing, Song Rui. UBGPM‑Markov based railway freight volume forecasting method. Dalian University of Technology Journal, 2014, 35(06):1‑5.