Bootstrap inference has become increasingly popular in econometrics due to advances in computer technology, allowing for the use of simulated distributions rather than asymptotic ones. This paper reviews bootstrap inference, including Monte Carlo tests, bootstrap tests, and bootstrap confidence intervals. While bootstrap methods often provide more accurate inferences than traditional asymptotic methods, they are not always reliable. The paper discusses the use of bootstrap methods in various contexts. For example, Monte Carlo tests, which are a type of bootstrap test, can be used to calculate exact P-values by simulating data. Bootstrap tests are particularly useful when the distribution of the test statistic is unknown or complex. However, they may not always work well, especially in cases with serial correlation, heteroskedasticity, or in simultaneous equations models. Bootstrap tests are generally more accurate than asymptotic tests in finite samples, especially when the test statistic is asymptotically pivotal. However, they can still have issues, such as overrejection or underrejection, depending on the model and data. The paper provides examples where bootstrap tests perform well, such as in models with exogenous or predetermined regressors and independent, identically distributed errors. However, in models with serial correlation, heteroskedasticity, or simultaneous equations, bootstrap tests may not always be reliable. The paper also discusses the use of different bootstrap methods, such as the sieve bootstrap and the block bootstrap, which are used to handle serial correlation in time series data. The wild bootstrap is another method used to handle heteroskedasticity, and it is shown to perform better in some cases than the pairs bootstrap. In simultaneous equations models, bootstrap methods are more complex because they must generate all endogenous variables, not just one. The paper illustrates that bootstrap methods can be less reliable in such models, especially when the error terms are homoskedastic. Overall, bootstrap inference is a valuable tool in econometrics, but it must be used with caution, as it is not always reliable in all situations. The paper highlights the importance of understanding the limitations of bootstrap methods and the need for careful application in different contexts.Bootstrap inference has become increasingly popular in econometrics due to advances in computer technology, allowing for the use of simulated distributions rather than asymptotic ones. This paper reviews bootstrap inference, including Monte Carlo tests, bootstrap tests, and bootstrap confidence intervals. While bootstrap methods often provide more accurate inferences than traditional asymptotic methods, they are not always reliable. The paper discusses the use of bootstrap methods in various contexts. For example, Monte Carlo tests, which are a type of bootstrap test, can be used to calculate exact P-values by simulating data. Bootstrap tests are particularly useful when the distribution of the test statistic is unknown or complex. However, they may not always work well, especially in cases with serial correlation, heteroskedasticity, or in simultaneous equations models. Bootstrap tests are generally more accurate than asymptotic tests in finite samples, especially when the test statistic is asymptotically pivotal. However, they can still have issues, such as overrejection or underrejection, depending on the model and data. The paper provides examples where bootstrap tests perform well, such as in models with exogenous or predetermined regressors and independent, identically distributed errors. However, in models with serial correlation, heteroskedasticity, or simultaneous equations, bootstrap tests may not always be reliable. The paper also discusses the use of different bootstrap methods, such as the sieve bootstrap and the block bootstrap, which are used to handle serial correlation in time series data. The wild bootstrap is another method used to handle heteroskedasticity, and it is shown to perform better in some cases than the pairs bootstrap. In simultaneous equations models, bootstrap methods are more complex because they must generate all endogenous variables, not just one. The paper illustrates that bootstrap methods can be less reliable in such models, especially when the error terms are homoskedastic. Overall, bootstrap inference is a valuable tool in econometrics, but it must be used with caution, as it is not always reliable in all situations. The paper highlights the importance of understanding the limitations of bootstrap methods and the need for careful application in different contexts.