Predictive Coding Highlighted in FHFA Case
Federal Housing Finance Agency v. HBSC North America Holdings Inc., et al., Case 1:11-cv-06189-DLC (S.D.N.Y. February 14, 2014) is an order regarding consolidated cases in New York where Plaintiff FHFA is suing multiple defendant banks, including Bank of America. As background, there was a similar action taking place in California that involved Countrywide (a subsidiary of Bank of America.) Defendants filed a Motion for Reconsideration regarding the court’s prior New York Order of January 8 that said any documents or electronic data produced in the California Countryside cases, but not the New York cases, could not be used for the New York cases. (Note that discovery and document production in these cases was long closed.)
Defendants objected to this Order as overbroad, as they argued they produced extensive discovery in the Countrywide case and that the Order should not preclude them from using the documents at trial. The court disagreed, plainly stating that Defendants may not use these documents not produced in discovery of the present cases at trial, as that would undermine the entire discovery process and “encourage trial by ambush.”
The court did carve out a narrow exception. If Defendants believe a witness has committed perjury at trial and could use a document to impeach the witness, they may bring this to the Court prior to the first day of jury selection.
Defendants also wished to use documents produced in the Countrywide litigation that show Plaintiff FHFA did not comply with its discovery obligations in New York. The Court rejected this request to use Countrywide documents to challenge the completeness of the discovery production in New York to avoid re-opening the entire issue of document production.
The Court offered an example of why it ruled this way. In the beginning of this litigation, the court allowed a Defendant to use predictive coding over objection. The Court viewed research that showed predictive coding (also called automated issue coding) had a better track record than human review, but neither process was perfect. The Court cited a publication that human review produced 25-80% of relevant documents, while predictive coding returned between 67-86%. The Court’s point was that no process is perfect, and it stated that had Defendants wished to challenge Plaintiff’s discovery production, they should have done so long before the present motion. “The Countrywide documents have not been produced in the New York Actions and they cannot be used as a springboard to reopen document discovery in these actions.”