谢谢！

我的论文现在建立了模型：

y=a1x1+a2x1*x2+a3x3+a4x4

其中x1是哑变量并可以用x5 x6 x7预测

现在认为x1具内生性，想用工具变量法来解决，但工具变量法的第一阶段好像不能是哑变量，特别x1和x2还有交叉项。

请问如何实现？下面是篇外文，他的处理方法我看不太懂，你能帮我看下吗？

This approach has three steps. In the first step, we predict auditor
quality as in Table 3. In steps 2 and 3, we use a standard instrumental approach, with the
predicted value obtained from step 1 as an instrument.  In our subsequent analysis we interact
BigSix （六大事务所）with past stock returns（上期股票收益）, we follow Wooldridge (2002) and used the demeaned value of
past returns.13 Wooldridge (2002) indicates that, under fairly general conditions, this procedure
provides efficient estimations. In addition, we are using the fitted probability of auditor quality
as an instrument, the effect of misspecification in auditor quality model is mitigated. We also
correct the standard errors for heteroskedasticity and serial correlation.