Previsão da Arrecadação de ICMS de São Paulo:Experimento com Econometria, Deep Learning e Foundation Model, sob Abordagens Top-Down e Bottom-Up
DOI:
https://doi.org/10.55532/1806-8944.2026.314Keywords:
Previsão; séries temporais; política tributária; aprendizado de máquina; econometria.Abstract
This article evaluates seven forecasting methodologies for the monthly ICMS revenue of the State of São Paulo under top-down and bottom-up strategies: SARIMA, SARIMAX in differences, SARIMAX in levels, ECM, LSTM, TimesFM and SCINet. The experiment uses rolling-origin windows, annual and monthly accuracy metrics, and ordinal and cardinal performance comparisons. Results show that models with exogenous variables dominate the ranking. Bottom-up ECM and SARIMAX deliver the best annual forecasts, while bottom-up SARIMAX also leads monthly accuracy; LSTM is the best-performing deep learning model. Overall, the evidence favors econometric specifications with economic covariates and indicates that revenue disaggregation by collection regime improves ICMS forecasting and may inform future IBS forecasting.
References
BONATO, A. S. F. Utilização de Modelos Não Lineares para Previsão do ICMS Paulista. 2023. 170 p. Dissertação (Dissertação (Mestrado em Economia)) — Fundação Getúlio Vargas/ Escola de Economia de São Paulo, São Paulo, 2023. Disponível em: https://repositorio.fgv.br/items/15a860b4-0458-4c68-b5aa-5f9b0789a5e4. Acesso em: 24/06/2026.
BOX, George E. P.; JENKINS, Gwilym M. Time Series Analysis: Forecasting and Control. San Francisco: Holden-Day, 1970.
BROCK, W.; DECHERT, W.; SCHEINKMAN, J. A test for independence based on the correlation dimension. SSRN Working Paper, 1987. Disponível em: https://doi.org/10.1080/07474939608800353. Acesso em: 24/06/2026.
DAS, A. et al. A decoder-only foundation model for time-series forecasting. In: PROCEEDINGS OF THE 41st INTERNATIONAL CONFERENCE ON MACHINE LEARNING. [S. l.]: PMLR, 2024. p. 10148-10167. Disponível em: https://proceedings.mlr.press/v235/das24c.html. Acesso em: 24 jun. 2026.
DICKEY, David A.; FULLER, Wayne A. Distribution of the estimators for autoregressive time series with a unit root. Journal of the American Statistical Association, [s. l.], v. 74, n. 366, p. 427-431, 1979. DOI: https://doi.org/10.1080/01621459.1979.10482531. Acesso em: 24/06/2026.
DORNELAS, A. F.; CAMPOS, L. D.; FIGUEIREDO, K. Modelos para previsão tributária utilizando redes neurais lstm. In: Anais do XIX Encontro Nacional de Inteligência Artificial e Computacional. SBC, 2022. p. 95–105. Disponível em: https://doi.org/10.5753/eniac.2022.227595. Acesso em: 24/06/2026.
DORNELLES, G. Z.; SCHWARTZER, F. R.; BRAATZ, J. Redes Neurais Aplicadas na Previsão de Receita de ICMS no Rio Grande do Sul. Textos para Discussão TE/RS Nº 19. Rio Grande do Sul, Brasil, 2022. Disponível em: https://tesouro.fazenda.rs.gov.br/upload/arquivos/202508/29094521-1643376127-artigo-modelo-redes-neurais-texto-de-discussao.pdf. Acesso em: 30/06/2026.
ENGLE, Robert F. Autoregressive conditional heteroscedasticity with estimates of the variance of United Kingdom inflation. Econometrica, v. 50, n. 4, p. 987–1007, 1982. DOI: 10.2307/1912773. Disponível em: https://doi.org/10.2307/1912773. Acesso em: 24/06/2026.
ENGLE, Robert F.; GRANGER, Clive W. J. Co-integration and Error Correction: Representation, Estimation, and Testing. Econometrica, [s. l.], v. 55, n. 2, p. 251-276, mar. 1987. Disponível em: https://doi.org/10.2307/1913236. Acesso em: 28 maio 2026.
ENGLE, R. F.; HENDRY, D. F.; RICHARD, J.-F. Exogeneity. Econometrica, [S. l.], v. 51, n. 2, p. 277–304, mar. 1983. DOI: 10.2307/1911990. Disponível em: https://doi.org/10.2307/1911990. Acesso em: 10 jun. 2026.
FONTENELE, S. V. P. Arrecadação Tributária: Modelos de Previsão para o Estado do Ceará. 2017. Dissertação (Dissertação (Mestrado Profissional em Economia do Setor Público – MESP)) — Programa de Pós-Graduação em Economia (CAEN), Universidade Federal do Ceará, Fortaleza, 2017.
FRIEDMAN, M. The use of ranks to avoid the assumption of normality implicit in the analysis of variance. Journal of the American Statistical Association, v. 32, n. 200, p. 675–701, 1937. DOI: https://doi.org/10.1080/01621459.1937.10503522. Acesso em: 24/06/2026.
HAMILTON, James D. Time Series Analysis. Princeton: Princeton University Press, 1994.
HAN, Z. et al. A review of deep learning models for time series prediction. IEEE Sensors Journal, v. 21, n. 6, p. 7833–7848, 2021. Disponível em: https://doi.org/10.1109/JSEN.2019.2923982. Acesso em: 24/06/2026.
HANSEN, P. R.; LUNDE, A.; NASON, J. M. The model confidence set. Econometrica, v. 79, n. 2, p. 453–497, 2011. DOI: https://doi.org/10.3982/ECTA5771.
HOCHREITER, S.; SCHMIDHUBER, J. Long short-term memory. Neural Computation, v. 9, n. 8, p. 1735–1780, 1997.
HYNDMAN, R. J.; KOEHLER, A. B. Another look at measures of forecast accuracy. International Journal of Forecasting, v. 22, n. 4, p. 679–688, 2006. DOI: https://doi.org/10.1016/j.ijforecast.2006.03.001. Acesso em: 24/06/2026.
HYNDMAN, R. J.; ATHANASOPOULOS, G. Forecasting: principles and practice. 3rd. ed. Melbourne, Australia: OTexts, 2021. Disponível em: OTexts.com/fpp3. Acesso em: 23 out. 2025.
JOHANSEN, Søren. Statistical analysis of cointegration vectors. Journal of Economic Dynamics and Control, [s. l.], v. 12, n. 2-3, p. 231-254, 1988. DOI: https://doi.org/10.1016/0165-1889(88)90041-3. Acesso em: 24/06/2026.
JOHANSEN, Søren. Estimation and hypothesis testing of cointegration vectors in Gaussian vector autoregressive models. Econometrica, [s. l.], v. 59, n. 6, p. 1551-1580, 1991. DOI: https://doi.org/10.2307/2938278. Acesso em: 24/06/2026.
JOHANSEN, S. Likelihood-based inference in cointegrated vector autoregressive models. Oxford: Oxford University Press, 1995.
JONES, D. R.; PERTTUNEN, C. D.; STUCKMAN, B. E. Efficient global optimization of expensive black-box functions. Journal of Global Optimization, Kluwer Academic Publishers, v. 13, n. 4, p. 455–492, 1998.
KINGMA, D. P.; BA, J. Adam: a method for stochastic optimization. In: INTERNATIONAL CONFERENCE ON LEARNING REPRESENTATIONS (ICLR), 3., 2015, San Diego. Proceedings [...]. San Diego: ICLR, 2015. p. 1–15. Disponível em: https://arxiv.org/abs/1412.6980. Acesso em: 30 jun. 2026.
KWIATKOWSKI, Denis; PHILLIPS, Peter C. B.; SCHMIDT, Peter; SHIN, Yongcheol. Testing the null hypothesis of stationarity against the alternative of a unit root: how sure are we that economic time series have a unit root? Journal of Econometrics, [s. l.], v. 54, n. 1-3, p. 159-178, 1992. DOI: https://doi.org/10.1016/0304-4076(92)90104-Y. Acesso em: 24/06/2026.
LI, W.; LAW, K. L. E. Deep learning models for time series forecasting: A review. IEEE Access, Piscataway, v. 12, p. 92306–92327, 2024. Disponível em: https://doi.org/10.1109/ACCESS.2024.3422528. Acesso em: 24/06/2026.
LIU, M. et al. SCINet: Time Series Modeling and Forecasting with Sample Convolution and Interaction. 2022. ArXiv. Disponível em: http://arxiv.org/abs/2106.09305v3. Acesso em: 8 set. 2025.
MASINI, R. P.; MEDEIROS, M. C.; MENDES, E. F. Machine learning advances for time series forecasting. Journal of Economic Surveys, v. 37, n. 1, p. 76–111, fev. 2023.
NEMENYI, P. Distribution-free multiple comparisons. 1963. Tese (Doutorado) — Princeton University, Princeton, 1963.
PHILLIPS, Peter C. B.; PERRON, Pierre. Testing for a unit root in time series regression. Biometrika, [s. l.], v. 75, n. 2, p. 335-346, 1988. DOI: https://doi.org/10.1093/biomet/75.2.335. Acesso em: 24/06/2026.
PHILLIPS, P. C. B.; HANSEN, B. E. Statistical inference in instrumental variables regression with I(1) processes. Review of Economic Studies, v. 57, n. 1, p. 99–125, 1990.
PHILLIPS, P. C. B. Optimal inference in cointegrated systems. Econometrica, v. 59, n. 2, p. 283–306, 1991.
POLITIS, D. N.; ROMANO, J. P. The stationary bootstrap. Journal of the American Statistical Association, v. 89, n. 428, p. 1303–1313, 1994. DOI: https://doi.org/10.1080/01621459.1994.10476870. Acesso em: 24/06/2026.
SAID, Said E.; DICKEY, David A. Testing for unit roots in autoregressive-moving average models of unknown order. Biometrika, [s. l.], v. 71, n. 3, p. 599-607, 1984. DOI: https://doi.org/10.1093/biomet/71.3.599. Acesso em: 24/06/2026.
SILVA, P.; FIGUEIREDO, K. Aprendizado profundo aplicado na previsão de receita tributária utilizando variáveis endógenas. In: Anais do XVII Encontro Nacional de Inteligência Artificial e Computacional. Porto Alegre, RS, Brasil: SBC, 2020. p. 414–425. ISSN 2763-9061. Disponível em: https://sol.sbc.org.br/index.php/eniac/article/view/12147. Acesso em: 24/06/2026.
STOCK, J. H. Asymptotic properties of least squares estimators of cointegrating vectors. Econometrica, v. 55, n. 5, p. 1035–1056, 1987.
WHITE, H. A heteroskedasticity-consistent covariance matrix estimator and a direct test for heteroskedasticity. Econometrica, v. 48, n. 4, p. 817–838, 1980. DOI: 10.2307/1912934. Disponível em: https://doi.org/10.2307/1912934. Acesso em: 24/06/2026.
Published
How to Cite
Issue
Section
License
Copyright (c) 2026 CADERNOS DE FINANÇAS PÚBLICAS
This work is licensed under a Creative Commons Attribution 4.0 International License.
The articles can be copied, since quoted the source.
