Forecasting financial asset prices has always been a challenging task due to a vast array of factor. Forecasting stock prices using econometric models is satisfactory when the data exhibits a certain statistical regularity, however, this regularity is not standard for this type of data. Considering that the advancement of computer technology, making them increasingly faster, the popularization of high-speed internet, the development of secure trading platforms, and the interconnection of markets have significantly increased the volume of transactions. With the increase in the number of transactions per unit of time, it became impossible for a human trader to keep up with this volume of buying and selling transactions of financial assets, thus giving rise to trading robots. In this chapter is made a comparision between traditional methods and machine learning methods to forecast time series

Being financed by third-party capital requires the companies to put up collateral or assets in guarantee, consisting of real estate, inventories, and accounts receivable that in turn depend on specific life cycles, among other aspects. The main object of this study is to analyze whether corporate debt is related to age and collateral. To do so, a sample of 194 public and private Brazilian companies was studied between 2010 and 2017. The findings indicate that more mature businesses have lower debt levels. In terms of the collateral variable and interactions between collateral and age, a negative relation was noted with financial leverage, contrary to what was expected. This fact indicates a possible lack of quality in collateral over time. Furthermore, it is noted that there is no directly proportional relationship between progression in age and collateral. The contribution of the study consists of analyzing the relationships between collateral and age in terms of the debt levels of public and private Brazilian businesses. The distinctions between these groups may throw light on organizations in the emerging countries in terms of how to handle financing decisions with financial and capital market institutions.

Forecasting returns in financial markets is notoriously challenging due to the resemblance of price changes to white noise. In this paper, we propose novel methods to address this challenge. Employing high-frequency Brazilian stock market data at one-minute granularity over a full year, we apply various statistical and machine learning algorithms, including Bidirectional Long Short-Term Memory (BiLSTM) with attention, Transformers, N-BEATS, N-HiTS, Convolutional Neural Networks (CNNs), and Temporal Convolutional Networks (TCNs) to predict changes in the price ratio of closely related stock pairs. Our findings indicate that a combination of reversion and machine learning-based forecasting methods yields the highest profit-per-trade. Additionally, by allowing the model to abstain from trading when the predicted magnitude of change is small, profits per trade can be further increased. Our proposed forecasting approach, utilizing a blend of methods, demonstrates superior accuracy compared to individual methods for high-frequency data.

Objetivo: verificar se a dependência serial de primeira ordem dos retornos diários do Ibovespa varia conforme o estresse financeiro global (VIX) e o volume negociado, e se essa variação resiste a controles do tamanho do retorno anterior, do calendário e da medida de volume. Métodos: 3.714 pregões (2011–2025); autorregressão de primeira ordem estimada em quatro regimes, definidos pelas medianas do ln VIX e do volume relativo defasados; erros-padrão HAC, bootstrap por blocos, medidas alternativas de volume e avaliação fora da amostra com janela expansiva. Resultados: a autocorrelação global é estatisticamente nula (φ = −0,078; p = 0,127), mas as inclinações diferem entre regimes (Wald χ²(3) = 10,98; p = 0,012). A diferença em diferenças do efeito do volume entre VIX alto e baixo é −0,305 (p = 0,006) e negativa em 18 das 19 estimativas distintas examinadas, mas deixa de ser significativa quando o regime de volume é depurado do valor absoluto do retorno anterior (p = 0,614), dessazonalizado (p = 0,141) ou medido pelo volume de futuros do Ibovespa (p = 0,920). Fora da amostra, o modelo por regimes não supera a média histórica (R² = −0,16%; Clark–West, p = 0,166). Conclusão: a heterogeneidade da dependência serial entre regimes é pequena, sensível à medida de volume e sem ganho preditivo fora da amostra; a modulação do efeito do volume pelo VIX é hipótese exploratória, e não achado estabelecido, e a evidência não é causal.