Forecasting household consumption in Nigeria: a comparison of machine learning and econometric models
DOI:
https://doi.org/10.64497/jssci.309Keywords:
household consumption forecasting, HFCE, machine learning, econometric models, XGBoost, expanding-window validation, NigeriaAbstract
Accurate short-term forecasts of household consumption are important for macroeconomic surveillance and policy planning, particularly in data-constrained emerging economies. This paper compares machine-learning and econometric approaches for forecasting quarterly household final consumption expenditure (HFCE) growth in Nigeria under a strict one-quarter-ahead information structure. The comparison covers simple forecasting benchmarks, autoregressive and ARIMA-based econometric specifications, penalised regression, tree-based machine learning and extreme gradient boosting. External predictors are lagged by one quarter, while model specification, feature selection and hyperparameter choice are determined exclusively through expanding-window validation. A nine-quarter holdout sample is reserved for final out-of-sample evaluation. Forecast performance is assessed using RMSE, MAE, sMAPE, MASE and RMSSE, complemented by small-sample-adjusted Diebold–Mariano–Harvey–Leybourne–Newbold tests, empirical validation-residual prediction intervals, leave-one-quarter-out sensitivity analysis and residual diagnostics. Among the validation-selected model-family representatives, XGBoost records the lowest RMSE (9.646) and MAE (7.772), followed by Gradient Boosting (10.591; 9.235), the classical AR(4) model (11.179; 10.231) and LASSO (18.216; 14.407). XGBoost also records the lowest sMAPE, MASE and RMSSE and remains ranked first by RMSE and MAE in all nine leave-one-quarter-out exercises. Formal predictive-accuracy tests show that XGBoost and Gradient Boosting significantly outperform the naïve benchmark under squared and absolute loss at the 5 per cent level, although their advantages over the historical-mean and four-quarter rolling-mean benchmarks are not statistically significant. SHAP analysis indicates that the validation-selected XGBoost model draws predictive information from lagged Fishing growth, historical HFCE growth, gross fixed capital formation and other sectoral indicators, while validation-fold analysis shows that predictor stability varies across estimation windows. The findings show that machine-learning models, particularly XGBoost, provide competitive gains in short-term HFCE forecasting, while conventional econometric models and simple benchmarks remain important comparators within a diversified and uncertainty-aware forecasting framework.
Downloads
References
[1] Y. Carrière-Swallow and F. Labbé, “Nowcasting with Google Trends in an emerging market,” Journal of Forecasting, vol. 32, no. 4, pp. 289–298, 2011, doi: 10.1002/for.1252. DOI: https://doi.org/10.1002/for.1252
[2] F. D’Amuri and J. Marcucci, “The predictive power of Google searches in forecasting US unemployment,” International Journal of Forecasting, vol. 33, no. 4, pp. 801–816, 2017, doi: 10.1016/j.ijforecast.2017.03.004. DOI: https://doi.org/10.1016/j.ijforecast.2017.03.004
[3] A. Fasulo, A. Guandalini, and M. D. Terribili, “Google Trends for nowcasting quarterly household consumption expenditure,” Rivista Italiana di Economia Demografia e Statistica, vol. 71, no. 4, pp. 5–14, 2017.
[4] G. Koop, S. McIntyre, J. Mitchell, and A. Poon, “Nowcasting ‘true’ monthly U.S. GDP during the pandemic,” National Institute Economic Review, vol. 256, no. 1, pp. 44–70, 2021, doi: 10.1017/nie.2021.8. DOI: https://doi.org/10.1017/nie.2021.8
[5] M. Marcellino and V. Sivec, “Nowcasting GDP growth in a small open economy,” National Institute Economic Review, vol. 256, no. 1, pp. 127–161, 2021, doi: 10.1017/nie.2021.13. DOI: https://doi.org/10.1017/nie.2021.13
[6] R. F. Niesert, J. A. Oorschot, C. P. Veldhuisen, K. Brons, and R.-J. Lange, “Can Google search data help predict macroeconomic series?,” International Journal of Forecasting, vol. 36, no. 3, pp. 1163–1172, 2020, doi: 10.1016/j.ijforecast.2018.12.006. DOI: https://doi.org/10.1016/j.ijforecast.2018.12.006
[7] D. Kohns and A. Bhattacharjee, “Nowcasting growth using Google Trends data: A Bayesian Structural Time Series model,” International Journal of Forecasting, vol. 39, no. 3, pp. 1384–1412, 2023, doi: 10.1016/j.ijforecast.2022.05.002. DOI: https://doi.org/10.1016/j.ijforecast.2022.05.002
[8] S. O. Akinjobi and T. O. Omooseti, “Forecasting Nigeria’s inflation components using machine learning and econometric models: A comparative analysis,” Research Square, 2025, doi: 10.21203/rs.3.rs-8160354/v1. DOI: https://doi.org/10.21203/rs.3.rs-8160354/v1
[9] D. C. Bartholomew, H. C. Iwu, and I. P. Ibemere, “A deep recurrent neural network approach to inflation forecasting in Nigeria: Evaluating the impact of feature selection methods,” Franklin Open, vol. 13, p. 100443, 2025. DOI: https://doi.org/10.1016/j.fraope.2025.100443
[10] J. F. Pfahler, “Exchange rate forecasting with advanced machine learning methods,” Journal of Risk and Financial Management, vol. 15, no. 2, pp. 1–17, 2022. DOI: https://doi.org/10.3390/jrfm15010002
[11] M. Qureshi, N. Ahmad, S. Ullah, and A. R. Mustafa, “Forecasting real exchange rate using artificial intelligence and time series models,” Heliyon, vol. 9, no. 5, p. e16335, 2023, doi: 10.1016/j.heliyon.2023.e16335. DOI: https://doi.org/10.1016/j.heliyon.2023.e16335
[12] C. P. Obite, A. Chukwu, D. C. Bartholomew, U. I. Nwosu, and G. E. Esiaba, “Classical and machine learning modeling of crude oil production in Nigeria,” Energy Reports, vol. 7, pp. 3497–3505, 2021. DOI: https://doi.org/10.1016/j.egyr.2021.06.005
[13] A. D. Gbadebo, “Leveraging machine learning for exchange rate prediction: A business and financial management perspective in Nigeria,” Reviews of Management Sciences, vol. 6, no. 2, pp. 37–52, 2024. DOI: https://doi.org/10.53909/rms.06.02.0254
[14] S. Smeekes and E. Wijler, “Macroeconomic forecasting using penalized regression methods,” International Journal of Forecasting, vol. 34, no. 3, pp. 408–430, 2018, doi: 10.1016/j.ijforecast.2018.01.001. DOI: https://doi.org/10.1016/j.ijforecast.2018.01.001
[15] A. B. Muhammad, O. O. Ishaq, B. B. Janet, M. A. Alhassan, S. Manzo, and S. Shehu, “A hybrid ExpAR-FIGARCH-ANN model for time series forecasting,” Journal of Statistical Sciences and Computational Intelligence, vol. 1, no. 4, pp. 283–293, 2025, doi: 10.64497/jssci.68. DOI: https://doi.org/10.64497/jssci.68
[16] A. U. Abubakar and I. Akeyede, “Bayesian regime-switching threshold autoregression for nonlinear dynamics and risk forecasting in the Nigerian stock exchange,” Journal of Statistical Sciences and Computational Intelligence, vol. 2, no. 2, pp. 313–325, 2026, doi: 10.64497/jssci.184. DOI: https://doi.org/10.64497/jssci.184
[17] P. O. Evans, M. O. Abifarin, and E. A. Ali, “Modeling and forecasting petroleum motor spirit (PMS) prices in Nigeria: An ARIMA-based time series analysis,” Journal of Statistical Sciences and Computational Intelligence, vol. 1, no. 4, pp. 485–494, 2025, doi: 10.64497/jssci.82. DOI: https://doi.org/10.64497/jssci.82
[18] M. Maaouane et al., “Using neural network modelling for estimation and forecasting of transport sector energy demand in developing countries,” Energy Conversion and Management, vol. 258, p. 115556, 2022. DOI: https://doi.org/10.1016/j.enconman.2022.115556
[19] U. U. Imoh and M. M. Rad, “Analysis and prediction of traffic conditions using machine learning models on Ikorodu Road in Lagos State, Nigeria,” Infrastructures, vol. 10, no. 5, p. 122, 2025. DOI: https://doi.org/10.3390/infrastructures10050122
[20] N. Ghosh, “Advancing GDP forecasting: Machine learning approaches across global, regional, and country-specific contexts,” 2025, doi: 10.22541/au.176402164.41325037/v1. DOI: https://doi.org/10.22541/au.176402164.41325037/v1
[21] X. Tang and Y. Xie, “Exchange rate forecasting: A deep learning framework combining adaptive signal decomposition and dynamic weight optimization,” International Journal of Financial Studies, vol. 13, no. 3, p. 151, 2025. DOI: https://doi.org/10.3390/ijfs13030151
[22] I. Elbatal, M. Sarwar, F. Jamal, M. Daniyal, Z. Hussain, and A. Ben Ghorbal, “Modelling on gross domestic product annual growth rate data by using time series, machine learning, and probability models,” Journal of Radiation Research and Applied Sciences, vol. 18, p. 101481, 2025. DOI: https://doi.org/10.1016/j.jrras.2025.101481
[23] S. A. Eshiemogie, P. P. Aielumoh, T. Okamkpa, M. C. Jude, L. Efe, A. N. Amenaghawon, H. Darmokoesoemo, and H. S. Kusuma, “A machine learning-supported framework for predicting Nigeria’s optimal energy storage and emission reduction potentials,” Renewable Energy Focus, vol. 53, p. 100677, 2025. DOI: https://doi.org/10.1016/j.ref.2024.100677
[24] T. Schmidt and S. Vosen, “Forecasting private consumption: Survey-based indicators vs. Google Trends,” Ruhr Economic Papers, no. 155, 2009. DOI: https://doi.org/10.2139/ssrn.1514369
[25] M. Simionescu and J. Cifuentes-Faura, “Can unemployment forecasts based on Google Trends help government design better policies? An investigation based on Spain and Portugal,” Journal of Policy Modeling, vol. 44, no. 1, pp. 1–21, 2022, doi: 10.1016/j.jpolmod.2021.09.011. DOI: https://doi.org/10.1016/j.jpolmod.2021.09.011
[26] M. Simionescu and A. G. Raišienė, “A bridge between sentiment indicators: What does Google Trends tell us about COVID-19 pandemic and employment expectations in the EU new member states?,” Technological Forecasting and Social Change, vol. 173, p. 121170, 2021, doi: 10.1016/j.techfore.2021.121170. DOI: https://doi.org/10.1016/j.techfore.2021.121170
[27] Y. Uematsu and S. Tanaka, “High-dimensional macroeconomic forecasting and variable selection via penalized regression,” The Econometrics Journal, vol. 22, no. 1, pp. 34–56, 2019, doi: 10.1111/ectj.12117. DOI: https://doi.org/10.1111/ectj.12117
[28] S. Tonini, F. Chiaromonte, and A. Giovannelli, “On the impact of serial dependence on penalized regression methods,” LEM Working Paper Series, no. 2022/21, Laboratory of Economics and Management, Sant’Anna School of Advanced Studies, Pisa, Italy, Jul. 2022.
[29] J. H. Ospina-Holguín and A. M. Padilla-Ospina, “Penalised regressions vs. autoregressive moving average models for forecasting inflation,” Económicas CUC, vol. 41, no. 1, pp. 65–80, 2020, doi: 10.17981/econcuc.41.1.2020.Econ.3. DOI: https://doi.org/10.17981/econcuc.41.1.2020.Econ.3
[30] M. Friedman, A Theory of the Consumption Function. Princeton, NJ, USA: Princeton University Press, 1957.
[31] F. Modigliani and R. Brumberg, “Utility analysis and the consumption function: An interpretation of cross-section data,” in Post-Keynesian Economics, K. K. Kurihara, Ed. New Brunswick, NJ, USA: Rutgers University Press, 1954, pp. 388–436.
[32] A. Deaton, “Saving and liquidity constraints,” Econometrica, vol. 59, no. 5, pp. 1221–1248, 1991. DOI: https://doi.org/10.2307/2938366
[33] C. D. Carroll, “A theory of the consumption function, with and without liquidity constraints,” Journal of Economic Perspectives, vol. 15, no. 3, pp. 23–45, 2001. DOI: https://doi.org/10.1257/jep.15.3.23
[34] J. B. Ajefu, “Income shocks, informal insurance mechanisms, and household consumption expenditure,” International Journal of Social Economics, vol. 44, no. 12, pp. 1818–1832, 2017, doi: 10.1108/IJSE-04-2015-0094. DOI: https://doi.org/10.1108/IJSE-04-2015-0094
[35] J. Animashaun and A. Wossink, “How do households cope during aggregate shocks? Evidence from the 2009–2015 oil crisis in Nigeria,” Resources Policy, vol. 95, Art. no. 105170, 2024, doi: 10.1016/j.resourpol.2024.105170. DOI: https://doi.org/10.1016/j.resourpol.2024.105170
Downloads
Published
How to Cite
Issue
Section
License
Copyright (c) 2026 Nura Aliyu Kabuga
This work is licensed under a Creative Commons Attribution 4.0 International License.
- Abstract 301
- PDF 88
