Assessing the Quality of 'Furfine-Based' Algorithms

Abstract

Standard-Nutzungsbedingungen:Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden.Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen.Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in www.econstor.euThis paper presents preliminary fi ndings and is being distributed to economists and other interested readers solely to stimulate discussion and elicit comments. AbstractTo conduct academic research on the federal funds (fed funds) market, historically one of the most important fi nancial markets in the U.S., some empirical economists have used market level measures published by the Markets Group at the Federal Reserve Bank of New York (FRBNY). To obtain more disaggregate data, some researchers have relied on a separate source of information: individual transactions inferred indirectly from an algorithm based on the work of Furfi ne (1999). To date, however, the accuracy of this algorithm has not been formally established. In this paper, we conduct a test aimed at assessing the ability of the algorithm to identify correctly individual overnight fed funds transactions conducted by two banks, which are among the most active in the fed funds market. The results are discouraging. We estimate the average type I and type II errors from 2007 to 2011 to be 81% and 23%, respectively. Furthermore, we argue that these errors i) apply to almost half of the algorithm's output, ii) introduce systematic biases, and iii) may not subside when the algorithm's output is aggregated. Our results therefore raise serious concerns about the appropriateness of using the algorithm's output to study the fed funds market. Because the FRBNY Markets Group relies on a different source of data than the algorithm output, our results have no bearing on their understanding of the fed funds market and their calculation of market level measures, including the effective fed funds rate.

Evaluating the Quality of Fed Funds Lending Estimates Produced from Fedwire Payments Data

“…(p. 5). We echo the caution of Armantier and Copeland (2012) in emphasizing that these estimates are not precise measures and their validity in any particular project should be interpreted only with a careful understanding of the process by which they are extracted from payments data and with an eye towards understanding the known weaknesses of the methodology. Table #3d: Bank #1 -Lend…”

“…However, until such time as improved data are collected and disseminated to researchers, we conclude that the estimates produced by the FRBNY algorithm are useful measures of overnight fed funds activity, specifically in that they are correlated with one measure of such activity. This approach has some advantages relative to the test proposed by Armantier and Copeland (2012). First, it allows the evaluation of the algorithm on multiple banks over time.…”

“…Interestingly, time fixed effects do not have significant explanatory power in explaining the difference between the Y-9C and algorithm estimated amount (Specification 10). Armantier and Copeland (2012) propose a test of algorithm estimates of fed funds transactions in which they examine in detail the payments of two large banks. Their methodology makes use, for each of two particular banks, of a bank-provided code which serves to "identify unambiguously the receive leg of every fed funds transaction" for the two banks' payments over the Fedwire Funds Service.…”

“…To be conservative, we believe that the estimates are best interpreted as measures of overnight interbank activity rather than fed funds activity specifically. We also compare the estimates provided by Armantier and Copeland (2012) to the Y-9C fed funds amounts. …”

“…Verifying the algorithm is complicated by definitional differences. Fed funds loans may refer very narrowly to unsecured overnight loans of immediately available reserves at banks (known as "Regulation D" fed funds, as defined by the Board of Governors of the Federal Reserve System, and as used in Armantier and Copeland (2012)). The Y-9C definition appears somewhat broader as it includes all overnight loans among banks that are not secured by securities.…”