Input Specificity and the Propagation of Idiosyncratic Shocks in Production Networks

Barrot, Jean-Noel, and Julien Sauvagnat, QJE (2004).

Question:

Do firm-level shocks propagate in production networks?

Methods:

  • Use natural disasters as the exogenous shock to suppliers
  • Compare firms whose suppliers are in disaster areas with firms who do not have any suppliers in disaster areas.
  • Use three measures of asset specificity:
    • Rauch’s (1999) classification of the extent to which goods are traded on markets
    • Supplier’s R&D expenditures
    • Supplier’s number of patents
  • Regress customer change in sales when a specific supplier suffers a shock
  • Regress (non-disaster) supplier change in sales when a customer has another specific supplier that suffers a shock.
  • Control for firm characteristics, number of suppliers, and fiscal-quarter fixed effects.
  • Take appropriate consideration for
    • Customers who are affected by the same disaster as their supplier
      • exclude observations where customer and supplier are less than 300 miles apart
    • the fact that customers may select their suppliers based on natural disaster risk
      • this would bias against finding any results
    • customers lose sales through some other channel – e.g., if their own customer base is in the same area as the supplier
      • find that there are not results when the customer-supplier link is inactive

Data:

  • Data on non-financial firms with headquarters in the U.S. between 1978-2013
  • Firm characteristics and location (county) from Compustat
  • Regulation SFAS 131 requires firms to report the names of customers accounting for more than 10% of segment sales (starting in 1978)
    • Supply chain from Compustat Segment data – match customer names to Compustat by string-matching and by hand
  • Disaster date and location data from SHELDUS (Spatial Hazard and Loss Database for the U.S.) at the University of South Carolina
  • Data on supplier specificity
    • Recreate Rauch’s (1999) measure of input specificity based on SIC-code
    • R&D data from Compustat
    • patent data from Kogan, et al. (2012), retrieved in turn from Google patents

Findings:

  • When a supplier’s location is hit by a natural disaster, customers’ sales growth falls by 2 to 3 percent and their equity value falls by 1%.
    • This effect is not observed when the supplier and customer do not have an active relationship at the time of disaster.
    • This effect is only observed when the supplier supplies a specific input.
    • $1 of lost supplier sales leads to $2.4 of lost customer sales – shocks propagate.
  • Suppliers who are not hit by a disaster, but which have a customer who has another supplier that is in a disaster zone, are also negatively affected.

Comments:

  • This paper looks at the effect of supplier shocks on customers and on other suppliers.  The authors also look at the impact of customer shocks on suppliers, but not at the effect of customer shocks on specific suppliers, which they should have done in a paper about asset specificity.  The position this as a supply-side story, but they do not fairly consider the demand-side.
  • Something else the authors could do to support their story is to look at inventory.  The basic story is that when a specific supplier is hit by a disaster, the supply of a specific input falls, so the (customer) firm has no choice but to cut production, leading to lost sales.  By effectively using sales as a measure of production, they assume that (1) inventory levels are fixed and (2) the relative sales price of the customer’s final good is fixed.  If, as I would expect, customers deplete inventory to make up for lost production, then the affect of the supplier shock is even bigger than this paper shows.
  • The authors probably want to look at lost sales to see the effect of shocks on firms’ bottom line.  If pre-shock inventories were optimal, than any decrease in inventory has a cost associated with it, as well.  My point is that there may be, and probably are, other consequences beyond lost sales.  How do the lost sales compare in magnitude to the declines in market cap for these companies?