January 16, 2026
Analysis of vehicle appreciation factors in Brazil during the Covid-19 pandemic
Author: Marcio Rubio Martins Zacheo — Advisor: Charles Gomes da Silva
Summary prepared by the ResumeAI tool, an artificial intelligence solution developed by Instituto Pecege focused on synthesis and writing.
This study analyzes the economic and productive factors that influenced the atypical appreciation of vehicles in Brazil during the Covid-19 pandemic, using statistical and Data Science models. The objective was to identify the factors behind the rise in prices of used cars (one year old) and provide a predictive model to assist in the formulation of public policies in future crises. The premise is that the inversion of the depreciation curve, an unprecedented phenomenon in the national market, resulted from a confluence of supply and demand shocks that require quantitative analysis.
The automotive sector is a pillar of the Brazilian economy, crucial for industrial growth and job creation. Data from the Brazilian Institute of Geography and Statistics (IBGE, 2022) position the sector as the third most important in the industrial production of Gross Domestic Product (GDP), with an annual volume of R$ 345 billion, corresponding to 5% of the national total, behind only oil/gas and mining. The automotive industry generates over 106,000 direct jobs and produces 1.8 million automobiles annually, according to the National Association of Automotive Vehicle Manufacturers (ANFAVEA, 2024).
The Covid-19 pandemic, however, drastically altered consumption patterns and production. In Brazil, the consumer behavior regarding automobiles was modified, as pointed out by the study by Junger et al. (2021). The use of public transport suffered a drop of up to 80% in the number of passengers in 2020, according to the Institute of Applied Economic Research (IPEA, 2021). This retraction led many individuals to seek individual vehicles, generating unexpected demand pressure on the market.
This new demand encountered a severely restricted supply. The production of new vehicles fell to historic levels, with only 2,000 units in April 2020, about 1% of the sector’s normal capacity (ANFAVEA, 2024). The drop was a consequence of disruptions in global supply chains, notably the shortage of semiconductors, an essential component for modern vehicles (SIA, 2023). The mismatch between the growing demand for individual transportation and the collapsed supply of new vehicles resulted in the appreciation of used vehicles, which, according to Schulz et al. (2023), registered an appreciation of up to 20% according to the Fipe Table.
Although specific studies have addressed the phenomenon, such as Souza (2022), who attributed the 18.47% appreciation in used vehicles to scarcity of components, and international reports that documented similar movements in the United States (Cox Automotive, 2021) and Europe (OECD, 2024), there is a gap in the national literature. An analysis that simultaneously articulates price data, macroeconomic variables, and industrial production indicators through quantitative approaches is lacking. This study fills this gap, using Data Science models to offer an in-depth understanding of the determinants of vehicle appreciation in Brazil and provide an analytical tool for managing future sectoral crises.
The adopted methodology is quantitative, with a descriptive and correlational focus, using a non-experimental approach. The analysis was conducted in Python, with libraries such as Pandas, Statsmodels, and Numpy. The database, with a monthly frequency from January 2018 to July 2025, was built from sources such as the Central Bank of Brazil, IBGE, ANFAVEA, Tabela Fipe, and Semiconductor Industry Association (SIA), consolidating variables such as Selic rate, IPCA, exchange rate, GDP, industrial production, and global semiconductor supply.
The object of analysis was the Volkswagen Polo Comfortline TSI, chosen because it was the best-selling hatch in the period and had not undergone significant technical changes, ensuring consistency in the historical price series of the FIPE Table (Fenabrave, 20–). The data underwent a treatment and standardization process (data wrangling). A crucial step was the application of lags to the exogenous variables, recognizing that economic shocks do not propagate instantaneously. Semiconductor supply shocks were tested with lags of 12 to 24 months, while the impact of credit was considered almost contemporaneous, with lags of 0 to 3 months.
The main analytical tool was the SARIMAX (Seasonal AutoRegressive Integrated Moving Average with eXogenous regressors) model. Its choice is justified by its ability to capture the temporal dynamics of a series, incorporating the effect of external variables and seasonal patterns. According to Hyndman and Athanasopoulos (2018), extensions of ARIMA such as SARIMAX are powerful tools for complex time series, integrating autoregressive (AR), moving average (MA), differencing (I) components, and external regressors (X). The literature, such as that from the Society of Actuaries (2013), reinforces that the inclusion of exogenous variables improves predictive accuracy in economic scenarios.
The superiority of SARIMAX over approaches like simple linear regressions (which ignore temporal dependence) lies in its balance between robustness and interpretability. Unlike more complex machine learning models, which may require more data and offer less transparency, SARIMAX allows for the clear isolation and quantification of the impact of external shocks. Shumway and Stoffer (2017) argue that, in economics, the inclusion of exogenous variables is fundamental to representing observed behavior. Model validation was ensured by rigorous diagnostic tests, such as Ljung-Box for autocorrelation, Jarque-Bera for normality, and ARCH for heteroscedasticity.
Cross-Correlation Analysis (CCF) identified the temporal relationship between the variables. One of the findings was a strong negative correlation (–0.85) between global microchip production and depreciation, with a 24-month lag. This shows that the drop in semiconductor supply was linked to the increase in used car prices, with a lapse that reflects the automotive supply chain cycle. The finding is in line with reports from S&P Global Mobility (2023) and Endurance Warranty (2025), which reported a 35% increase in used car prices in the US.
Similarly, the CCF revealed a negative correlation of –0.86 between the Selic rate and depreciation, with a 14-month lag. This indicates that lower interest rates stimulate demand for financing, pushing prices up. The subsequent increase in the Selic rate tends to cool the market. The result corroborates analyses such as that of TradeMap (2022), which associated the 22% appreciation of used cars in 2021 with the low interest rate scenario. Based on the CCF, the variables with the greatest relevance were selected, following the guidance of Box et al. (2015) and Fávero and Belfiore (2017) on the importance of parsimony.
The application of the SARIMAX model began with a reduced specification, containing the Selic Rate (14-month lag) and Microchip Production (18-month lag). Both coefficients were negative and statistically significant at 5%. The economic interpretation is that the fall in Selic stimulates credit, while the shortage of semiconductors restricts the supply of new ones, sustaining the prices of used ones. Diagnostic tests confirmed the good specification of the model (p-value of 0.95 in the Ljung-Box test).
Next, a complete model was estimated, including Unemployment Rate, Consumer Confidence Index (ICC), IPCA, and Exchange Rate. In this specification, only Selic (14-month lag) and Exchange Rate (9-month lag) maintained strong significance. The relevance of the exchange rate highlights the influence of external costs. The microchips variable lost significance, suggesting an overlap of effects. The AR(1) and MA(1) terms did not prove relevant, indicating that the dynamics were predominantly explained by the exogenous variables.
To investigate the loss of significance of the chips variable, a multicollinearity test (VIF) revealed severe collinearity between Selic (VIF = 27.68) and Unemployment (VIF = 22.17), and moderate for ICC (VIF = 6.66). The Chips (VIF = 3.01), Exchange Rate (VIF = 2.49), and IPCA (VIF = 1.90) variables did not show concerning collinearity. According to Stock and Watson (2020), multicollinearity inflates standard errors, making it difficult to identify the isolated effect. The analysis suggests that the loss of significance of the chips was not due to a statistical problem, but to an overlap of its effect with the exchange rate, which also captures international supply chain shocks (Baldwin & Freeman, 2022).
A parsimonious and robust final model was chosen, with the three variables representing the main channels of influence: Selic (monetary policy), Chips (supply constraints), and Exchange Rate (international costs). This specification presented low VIFs for all variables (Selic = 2.87; Chips = 2.64; Exchange Rate = 1.42), confirming the absence of severe multicollinearity. This model was then submitted to accuracy evaluation, according to best econometric practices (Hyndman & Athanasopoulos, 2018).
The final SARIMAX (1,1,1) model, with the exogenous variables Selic (14-month lag), Chips (18-month lag), and Exchange Rate (9-month lag), presented satisfactory predictive and statistical performance. Selic remained the most robust variable, with significance at the 1% level. The Chips and Exchange Rate variables, although of marginal relevance, proved theoretically coherent, especially considering that exchange rate fluctuations amplify the supply shocks of imported inputs such as semiconductors (Gaspar, Hoff & Mano, 2024). The diagnostic tests again validated the model specification.
In the out-of-sample accuracy evaluation, over a 12-month horizon, the results were remarkable. The model achieved a Mean Absolute Error (MAE) of 0.0202, a Root Mean Squared Error (RMSE) of 0.0243, and a Symmetric Mean Absolute Percentage Error (sMAPE) of only 2.21%. This latter indicator translates to a predictive accuracy of 97.8%. Although the high result may reflect the stability of the test period, the model’s theoretical robustness and the absence of diagnostic issues suggest that this is not overfitting, but rather a successful specification.
The study demonstrated the complexity of automotive depreciation behavior during the pandemic, shaped by the interaction between macroeconomic factors and supply shocks. The SARIMAX model proved to be an effective tool for quantifying the lagged impact of crucial determinants. The analysis culminated in the definition of a parsimonious and robust model, based on three variables: Selic rate, microchip supply, and exchange rate. The Selic emerged as the factor with the greatest statistical robustness, while the supply variables, Chips and Exchange Rate, complemented the model with theoretical relevance. The model’s predictive performance, with 97.8% accuracy, reinforces its viability as an analytical tool for industry managers and public policymakers. The research demonstrates the potential of integrating data science and economic analysis to build interpretable models with high predictive power. It is concluded that the objective was achieved: it was demonstrated that a parsimonious SARIMAX model, incorporating the Selic rate, the supply of semiconductors, and the exchange rate with adequate lags, is capable of explaining and predicting with high accuracy the atypical appreciation of vehicles in Brazil during the pandemic period.
References:
ARAÚJO, Iara Soares de; BRANDÃO, Viviane Bernadeth Gandra. Trabalho e renda no contexto da pandemia de COVID-19 no Brasil. Revista Prâksis, Novo Hamburgo, v. 18, n. 2, p. 96-111, May/Aug. 2021. Available at: . Accessed: Nov. 5, 2024.
ASSOCIAÇÃO NACIONAL DOS FABRICANTES DE VEÍCULOS AUTOMOTORES – ANFAVEA. Edições em Excel. Available at: . Accessed: Nov. 2, 2024.
BALDWIN, R.; FREEMAN, R. Supply chain contagion waves: thinking ahead on manufacturing ‘contagion and reinfection’ from the COVID-19 shock. VoxEU, 2020.
BANCO CENTRAL DO BRASIL. Taxa Selic: séries temporais. Available at: https://www3. bcb. gov. br/sgspub/localizarseries/localizarSeries. do. Accessed: Aug. 27, 2025.
BOX, George E. P.; JENKINS, Gwilym M.; REINSEL, Gregory C.; LJUNG, Greta M. Time series analysis: forecasting and control. 5th ed. Hoboken: Wiley, 2015.
COX AUTOMOTIVE. Kelley Blue Book: average new-vehicle prices up nearly 10% year-over-year in August 2021. 2021. Available at: https://www. coxautoinc. com/wp-content/uploads/2021/09/KBB-August-2021-Average-Transaction-Prices. pdf. Accessed: Sep. 27, 2025.
ENDURANCE WARRANTY. Microchip shortage update: is the automotive industry still taking a hit? Jul. 2025. Available at: https://www. endurancewarranty. com/learning-center/news/microchip-shortage-update-is-the-automotive-industry-still-taking-a-hit/.
FÁVERO, Luiz Paulo; BELFIORE, Patrícia. Manual de análise de dados: estatística e modelagem multivariada com Excel®, SPSS® e Stata®. 2nd ed. Rio de Janeiro: Elsevier, 2017.
FEDERAÇÃO NACIONAL DA DISTRIBUIÇÃO DE VEÍCULOS AUTOMOTORES – FENABRAVE. Emplacamentos. [S. l.], [20–]. Available at: https://www. fenabrave. org. br/portalv2/Conteudo/emplacamentos. Accessed: Sep. 9, 2025.
FIPE – FUNDAÇÃO INSTITUTO DE PESQUISAS ECONÔMICAS. Índice de preços de veículos. Available at: https://veiculos. fipe. org. br. Accessed: Nov. 2, 2024.
GASPAR, Vitor; HOFF, Pieter; MANO, Rodrigo. Global supply chain pressures, inflation, and implications for monetary policy. VoxEU, Mar. 6, 2024.
HYNDMAN, R. J.; ATHANASOPOULOS, G. Forecasting: principles and practice. 2nd ed. Melbourne: OTexts, 2018. Available at: https://otexts. com/fpp2/. Accessed: Sep. 3, 2025.
INSTITUTO BRASILEIRO DE GEOGRAFIA E ESTATÍSTICA – IBGE. Pesquisa industrial anual – produto 2022. Available at: . Accessed: Nov. 4, 2024.
INSTITUTO DE PESQUISA ECONÔMICA APLICADA – IPEA. Tendências e desigualdades da mobilidade urbana no Brasil I: o uso do transporte coletivo e individual. Brasília; Rio de Janeiro: Ipea, 2021. (Texto para discussão). Available at: . Accessed: Nov. 5, 2024.
JUNGER, Alex Paubel; MESQUITA, Francisco Vinícius de Sousa; GONÇALVES, Ingrid Sequeira; OLIVEIRA, Mayara dos Santos; LIMA, Stella Santana de; OLIVEIRA, Victor Inácio de. Análise do perfil comportamental do consumidor de automóveis diante do cenário da COVID-19. Revista de Casos e Consultoria, v. 12, n. 1, e25632, 2021. Available at: https://doi. org/10.1590/S2237-7417.2021. e25632. Accessed: Nov. 5, 2024.
MAJKA, M. ARIMAX: time series forecasting with external variables. ResearchGate, 2024. Available at: https://www. researchgate. net/publication/384196976ARIMAXTimeSeriesForecastingwithExternalVariables. Accessed: Sep. 3, 2025.
OECD. The future of the automotive value chain. Paris: OECD Publishing, 2024. Available at: https://www. oecd. org/content/dam/oecd/en/publications/reports/2024/11/the-future-of-the-automotive-value-chain91460165/cb730d65-en. pdf. Accessed: Sep. 27, 2025.
S&P GLOBAL MOBILITY. The semiconductor shortage is mostly over for the auto industry. Jul. 2023. Available at: https://www. spglobal. com/mobility/en/research-analysis/the-semiconductor-shortage-is-mostly-over-for-the-auto-industry. html.
SCHULZ, Aleciane; ROCHA, Jéssica Silva; TRABACH, Maurílio Polak; ALMEIDA, Jóici Schaffer Soares de; ESTEVÃO, Simone Batista Fernandes; ZANONI, Helenize Espíndula Rossi Coser; PONATH, Aramis Monteiro; FURLANI, Cleidir José. Impacto da pandemia (COVID-19) na valorização de carros novos e seminovos no período compreendido entre 2020 a 2022. Revista Interdisciplinar da FARESE, v. 4, 2023. Available at: . Accessed: Nov. 5, 2024.
SEMICONDUCTOR INDUSTRY ASSOCIATION. Market data. Available at: https://www. semiconductors. org. Accessed: Aug. 24, 2025.
SHUMWAY, R. H.; STOFFER, D. S. Time series analysis and its applications: with R examples. 4th ed. Cham: Springer, 2017.
SOCIETY OF ACTUARIES. Forecasting long-term disability incidence rates using ARIMA and ARIMAX models. Schaumburg, IL: SOA, 2013. Available at: https://www. soa. org/globalassets/assets/files/research/projects/research-2013-arima-arimax-ben-appl-rates. pdf. Accessed: Sep. 3, 2025.
SOUZA, Roberto Capelari de. Analysis of the reference value of passenger vehicles during a period of the COVID-19 pandemic. 2022. 65 f. Conclusion Work (Undergraduate in Automotive Engineering) – Joinville Technology Center, Federal University of Santa Catarina, Joinville, 2022. Available at: https://repositorio. ufsc. br/handle/123456789/232630. Accessed: Sep. 1, 2025.
STOCK, J. H.; WATSON, M. W. Introduction to econometrics. 4th ed. New York: Pearson, 2020.
TRADEMAP. Used cars appreciate in the pandemic, but high prices are numbered. Agência TradeMap, Feb. 17, 2022. Available at: https://trademap. com. br/agencia/brasil/carros-usados-se-valorizam-na-pandemia-mas-precos-altos-estao-com-os-dias-contados.
Executive summary from the Final Course Work of the Specialization in Data Science and Analytics of the MBA USP/Esalq
Learn more about the course; click here:
