<p>Investors often exhibit hesitative behavior within the spectrum between their profit-taking and stop-loss thresholds, which makes the analysis of investment behavior extremely complicated. In order to describe this complex phenomenon, this paper introduces a <i>p</i>th-order double-banded-threshold mixture autoregressive (DBTMAR(<i>p</i>)) model, which cleverly utilizes a banded threshold structure. It can not only describe the behavior of investors buying and selling stocks, but also fully consider a vague state in which investors are hesitant to sell stocks. We solve the parameter estimation problem using the conditional least squares method and obtain the asymptotic properties of the estimators, including threshold parameters. Finally, an application with the real gross domestic product data of growth rate in the United States is provided.</p>