Threshold Events and Identication: A Study of Cash Shortfalls Bakke and Whited, published in the Journal of Finance in June 2012
Introduction The paper combines three objectives 1 Provide general guidelines on how one can use (and cannot use) threshold events to identify causal relationships Discussion framed in terms of regression discontinuity design 2 Revisit the results in Rauh (JF, 2006), who nds that rms cut capital expenditures 70 cents for every dollar of mandatory contributions to pension plans Rauh makes use of discontinuities in a regression analysis (more on this later) 3 Determine how exactly rms respond to cash shortfalls
Threshold Events & RDD Threshold event: discrete event or treatment occurs when an observable continuous variable passes a known threshold RDD: idea is that observations (rms) just above and below the threshold can be thought of close-to-randomly assigned to treatment Causal eect of treatment: calculate average dierences between outcome variable of these two groups of rms More formally: estimate y i1 y i0, where only one of these variables is observed for each rm focus on average eects of treatment over sub-populations rather than on individual eects
More on RDD RDD comes in two avors: sharp and fuzzy Sharp RDD: rms are assigned to treatment solely on the basis of an observed, continuous measure s, called selection variable Required: The selection variable has a positive density in a neighborhood of the cuto s rules out manipulation of the selection variable no bunching of observations on one side of the threshold Firms are not allowed to be able to manipulate treatment (random assignment...) RDD is a technique with strong internal (near the threshold) and weak external (far from the threshold) validity
Threshold Events & Pension Plans (Rauh, JF 2006) (i) Threshold event: mandatory contributions that rms must make to their dened benet plans If plan is underfunded (plan assets < plan liabilities), rms are required to make contributions Mandatory pension contributions are a discontinuous and kinked function of the funding gap Required: limitations on rms' ability to manipulate the funding gap In fact, this setting provides more than one threshold event (e.g., 10% underfunding, 20% underfunding)
Threshold Events & Pension Plans (Rauh, JF 2006) (ii) Rauh (2006) argues that the pension fund setting can identify independent variation in cash ow Even though pension funding status is endogenously determined with investment opportunities, the rule that relates mandatory contributions to funding status contains discontinuities and kinks Hence, as long as one controls for funding status (regression of investment on proxies for investment opportunities), mandatory contributions are uncorrelated with investment opportunities (conditional on funding status) Required: relation between investment opportunities does not contain the same kinks and discontinuities as the relation between mandatory contributions and funding status
Threshold Events & Pension Plans (Rauh, JF 2006) (iii) I it CF A i,t 1 = a i + a t + b 1 Q i,t 1 + b it MC 2 A i,t 1 + b 3 A i,t 1 + b it 4 A i,t 1 + u it I it is capital expenditures, CF it is non-pension cash ow, FS it is funding status, MC it is mandatory contribution FS it Identication strategy based on the idea that investment opportunities do not jump down at the point of underfunding (or change slope when mandatory contribution function changes slope) Implication suggested by Rauh: Investment opportunities cannot be correlated with mandatory contributions, conditional on funding status BUT: Although kinks and jumps clearly diminish any correlation, they do not necessarily set it to zero
The results revisited Full sample: Coecient on mandatory contributions is negative and signicant, as it was found in Rauh (2006) It is the observations that are 80% to 90% funded that are driving most of the results Strong sensitivity of investment to mandatory contributions stems from heavily underfunded rms that constitute a small fraction of the sample and that are dierent from the rest of the sample control group diers from treated group Important in these types of analysis: use both informal visual and formal statistical analysis
Formal statistical analysis
Visual analysis
RDD & Pension Plans It is important to examine whether treated rms (with large mandatory contributions) are dierent from control rms (with small or zero mandatory contributions) Group that is less than 90% funded diers systematically from the rest of the sample less investment, lower Q, 25% reduction in cash ow, 33% reduction in size, 50% reduction in dividends, 67% reduction in earnings Hints at potentially large dierences in unobservables This lends credence to the hypothesis that over the whole sample these dierences have as much a role to play in the determination of investment as do mandatory contributions RDD as an alternative to regression model: What happens at threshold points?
80%, 90%, and 100% funding points as threshold events
Eects on R&D, Receivables, Inventories, Employment, but not on Investment
Conclusion 1 New guidance on how to use threshold events as identication strategy 2 Evidence in Rauh (2006) stems from a small group of heavily underfunded rms which dier sharply from the rest of the sample 3 Authors nd causal eects of mandatory contributions on R&D, receivables, inventories and employment, but not on investment