Identification concerns

To study how M&A activity relates to macro labor trends, we need to consider all M&As irrespective of whether technology adoption was part of the ex-ante incentive to pursue the M&A or not. However, to conclude that M&As act as a catalyst for the adoption of labor saving technology, we need to address the alternative interpretation that an omitted variable (e.g. industry shock) may be driving both M&A activity and the associated occupational changes we document in the data. Our analysis allows us to absorb variation in industry and local conditions by controlling for time- varying industry and state fixed effects. In this section, we provide further evidence to mitigate an

omitted variable concern.

First, following the approach in Seru (2014) and Malmendier, Opp, and Saidi (2016), we con- sider a sample of M&A deals that were announced but subsequently cancelled for reasons exoge- nous to the targets labor needs. To identify this sample, we start with all M&A deals announced over the 2001-2007 period that were subsequently withdrawn. We then read Factiva news articles explaining the reasons for the cancellation and retain a sample of deals where the M&A was ei- ther blocked by regulators, typically for anti-trust concerns, or because the acquirer was acquired ex-post and had to withdraw the deal. This leaves us with a small sample of deals cancelled for reasons exogenous to the target’s labor demand.19 We are able to identify 33 establishments in the OES survey data with cancelled M&A deals and this forms our ‘pseudo treated’ group. Follow- ing the same matching procedure as described in Section 2.2, we create a control sample which excludes establishments involved in M&As over our sample period.

Table 2.8 repeats the specification in column 3, Table 2.2 controlling for establishment and industry times year fixed effects.20 _{We consider all our main dependent variables using this sample}

of ‘pseudo-treated’ deals and their matched control establishments. Across all our measures, we cannot replicate the same pattern as in our baseline results. In fact, all coefficients always take the opposite sign from what our hypotheses predict. These findings thus reinforce the notion that our difference-in-difference results capture the effect of M&As and not of some other confounding variables, as such omitted variables should impact target firms associated with completed M&As and the cancelled M&As in our sample equally.

Second, we perform estimations within establishments, absorbing any time-varying shocks at the establishment level that could be driving our results. To include establishment-year fixed

19_{The other most common reasons stated for why deals get cancelled include: the management of the acquirer or}

the target rejecting the deal; disagreement on the price; changes in market or industry conditions; or bad news being revealed for the target. However, these reasons are arguably not exogenous to the target’s labor demand and therefore we choose not to consider them.

effects, we need variation within establishment-year. To this end, we separate routine and non- routine employment. We then test our predictions regarding employment and wage outcomes post-M&A looking specifically at routine occupations, occupations that are known to be dispro- portionately impacted by labor-saving technology, while controlling for changes at non-routine occupations at the same establishment. Specifically, we defineRoutineto take a value of one for routine employment in a given establishment and 0 for non-routine employment. We then interact Routine with Postt ·M&Ai and estimate the effect of the M&A on routine occupations within

establishments in a triple differences specification.

In Table 2.9, we show a greater reduction in the share of employment (column 1) for routine (as opposed to non-routine) occupations in treated establishments following the acquisitions as compared to control establishments. These results suggest lower demand for tasks substitutable by technology in M&A targets— a prediction unique to our technology adoption hypothesis— which is estimated after fully controlling for any contamporaneous shocks at the establishment level that could be driving changes in employment or wages. We estimate a decline in employment in routine workers, relative to non-routine workers of 8.5%. Comparing this estimate to the estimated decline of 4.4% in Table 2.2, column 1, using all employees, suggests that non-routine workers gain in employment post-M&A adding to our earlier argument that the impact of an M&A on the target firms employees is conditional on the type of worker.

Likewise, we reported in Table 2.4 that wages, on average, increase. In column 2, we examine what is the effect on routine workers’ wages following the acquisition. We repeat the specification in column 1, except we additionally control for the share of employment by occupation type to control for the concurrent occupation employment changes that take place at the establishment. We show that wages for routine workers falls by -3.9%, suggesting differential wage results conditional on the occupational type.