Does investment-relevant information leak from the Fed? A new Stigler Center working paper finds surprising and novel evidence in a public dataset.



The Federal Reserve Building In Washington, DC

One of the more dismal aspects of the dismal science is the trade-off between the richness of data and one’s freedom to work with them. I exploit a notable exception: the New York City taxi regulator’s publication of over a billion anonymous yellow taxi pick-up and drop-off records back to 2009. In a new Stigler Center paperI assess whether variation in meetings between insiders of the Federal Reserve Bank of New York and major US commercial banks implied by the ride data is consistent with the hypothesis in Cieślak, Morse, and Vissing-Jørgensen (2016) that there has been systematic information leakage from the Federal Reserve. While the taxi data cannot provide direct evidence of leakage, they can shed light on opportunities for even unwitting information flow. 


The possibility that interactions with Federal Reserve (Fed) insiders can yield even accidental insights into the institution has been acknowledged both within and outside of the Federal Reserve System. For instance, a senior Federal Reserve official noted during a monetary-policy meeting in Washington, DC, that “…it can happen pretty innocently when an experienced reporter lures one into revealing things that end up crossing the line.” Information might also flow to commercial bankers. The CEO of a major asset manager is reported as stating that insights into the Federal Reserve could be obtained from questions that the central bank posed to members of the private sector. Cieślak, Morse, and Vissing-Jørgensen (2016) present evidence that the equity premium has largely been earned in the weeks of Fed monetary-policy meetings and, drawing on a corpus of anecdotes, hypothesize that systematic unofficial Fed communication around these monetary-policy meetings is responsible.


The taxi data can shed light on one small sliver of Federal Reserve interactions: face-to-face meetings with individuals associated with the Federal Reserve Bank of New York (New York Fed, FRBNY). Insiders of the New York Fed and commercial banks can take black cars, the subway, and other modes of transportation, so I will obtain a lower bound. While monetary policy is primarily the purview of the Board of Governors in Washington, DC, the New York Fed plays an important role. It houses the Fed’s trading desk and supports decision-making by, for example, providing economic briefings. In addition to its actions as a regulator, the New York Fed regularly communicates with commercial financial institutions to obtain market commentary pertinent to monetary policy.


If one accepts the proposition that the Federal Reserve’s monetary-policy meetings in Washington, DC, impact interactions between insiders of the New York Fed and major commercial banks much more than interactions involving the businesses and residences around the New York Fed and those banks, one may identify significant changes in rides with significant changes in meetings relevant to this study. I cannot conclusively demonstrate a link between rides and face-to-face meetings, but evidence that individuals are in very close proximity to each other more often around FOMC meetings would complement more indirect evidence of regular informal communication presented in the academic literature. Given my limited observations, the magnitudes of the changes in meeting counts that I find should be viewed as lower bounds


The basic empirical question that I ask is how many more interactions there are between insiders of the New York Fed and major commercial banks around a particularly important monetary-policy meeting—an FOMC meeting—than one would expect given the time of the day, the day of the week, the year-month (e.g. January 2018), and the overall volume of contemporaneous taxi rides in Manhattan. It is at FOMC meetings that policymakers make decisions regarding the federal funds target range and unconventional monetary policy, and the announcements that follow them are eagerly anticipated and closely scrutinized by participants in financial markets.


In my assessment of how interactions vary around these meetings, I use taxi rides between the vicinities of the New York Fed’s and major commercial banks’ buildings as indicators of meetings at those institutions, and I use coincidental drop-offs of passengers picked up around those institutions as indicators of offsite meetings. Under the baseline definition, coincidental drop-offs are mapped to the same block, occur within 10 minutes of each other, and are separated by no more than a quarter of the length of the short side of a midtown block, 66 feet, along the North-South and East-West axes. Almost surely many direct rides and coincidental drop-offs are unrelated to the institutions and interactions of interest, but statistical methods can cut through much of this noise. To maximize the share of rides with relevant passengers, I focus only on the New York Fed’s main premises and on buildings in Manhattan where particularly large American banks are headquartered or where their investment-banking, investment-management, or global-markets divisions are headquartered. 


I first examine whether the taxi data are informative about commercial bankers’ movements. Goldman Sachs’s relocation to its new headquarters provides a clean test. I find a very striking and significant increase in rides between the new headquarters and the other major commercial banks that is coincidental with the move, and I find the mirror image for the old headquarters, though the changes in ride volumes are not the same magnitude. The taxi data will provide at most partial coverage of the movements between the institutions, but this is clear evidence that I am capturing some of them. 


To assess whether the taxi data reflect the conduct of New York Fed business, I examine patterns of rides between the New York Fed and major commercial banks around key dates in the passage of Dodd-Frank. The New York Fed is an important regulator, and Dodd-Frank is the most significant reform of American financial regulation in many years. One would consequently expect meetings between the New York Fed and the banks that it regulates or between New York Fed staff based at its headquarters and those embedded at the institutions that it regulates. I find that rides between the New York Fed and the commercial banks are significantly elevated around the passage of Dodd-Frank, and the weekday before each of the final milestones—the filing of the conference report, its agreement in the House and in the Senate, and the bill’s signing into law—is at least in the top 99.7th percentile of daily ride volumes over my sample.


The most obvious starting point for this paper’s investigation of interactions between insiders of the New York Fed and major commercial banks around FOMC meetings is the set of direct rides between them.


In the interest of thoroughness and transparency, I break a day into short windows, and I estimate for each of them the change in rides on each of what are generally the weekdays from a week before the typical start of an FOMC meeting through a week after the announcement. I observe a striking increase in rides from the commercial banks to the New York Fed soon after a communications blackout is lifted from Fed staff. The blackout ends at midnight the day after the Federal Reserve announces its monetary-policy decisions, and I find roughly an extra ride every other meeting between 1:00 a.m. and 4:00 a.m., an increase of about 100 percent over what one would expect over that window. This amounts to between 3 and 4 extra rides per year and, in light of the partial coverage of interactions, is likely a lower bound. If I had examined only that window, there would be less than a one-in-a-thousand chance of observing an increase that large by chance. If I account for the examination of hundreds of windows in a conservative manner, the probability of an increase that high is still below 1 percent. A variety of changes to my modeling decisions yield similar results.


I repeat the exercise for coincidental drop-offs and find that rides are consistently elevated around noon from what is typically the first day of the FOMC meeting through the following week. Individually, the changes over those short, single-day windows are too likely to be by chance for me to make any strong conclusions. Nevertheless, evidence suggestive of changes in lunchtime meetings during a span when the literature argues that there is informal communication motivates further analysis. A standard result in statistics is that the more observations one has, the less uncertainty there is around estimates, and the less likely one is to ascribe a finding to chance incorrectly. While I cannot change the number of FOMC meetings during my sample, I can pool days around monetary-policy meetings together, going from, for example, 48 observations for the day of a meeting and 47 for the day afterwards to 95 observations for the set of those two days. 


I employ a data-mining exercise to assess whether lunchtime meetings are elevated around FOMC meetings. For each of a set of intraday windows between 11:00 and 14:00, I estimate the change in coincidences over each contiguous span of weekdays from a week before the typical start of an FOMC meeting through a week after the announcement. My best guess for the hours and days during which rides are elevated balances the estimated increase in rides and a measure of the likelihood that the result is purely by chance. While there is some minor variation depending on modeling choices, I find elevated rides between roughly the first day of the FOMC meeting through the following week. I estimate slightly over one extra lunchtime coincidence for each FOMC meeting, an increase of about 50 percent. If I do not account for the data mining, the likelihood of an increase as large is less than one in a thousand. When I account for the inspection of hundreds of specifications in a conservative manner, the probability that an increase that large is by chance is less than 5 percent.


Further examinations provide additional support both for an increase in planned lunchtime meetings and for the exceptionality of the period around an FOMC meeting. Coincidences largely occur in dining and shopping areas like the Meatpacking District, SoHo, and TriBeCa. While the number of lunchtime coincidental drop-offs in the high-density neighborhoods where financial institutions are concentrated might raise concerns that the increase reflects unrelated drop-offs in the course of official duties, I find that most of the increase occurs away from them.


Another concern is that after support work for a FOMC meeting is complete, New York Fed staff might go to lunch together and simply have similar taste to that of commercial bankers. However, I find that the increase is mainly driven by the coincidental drop-offs of taxis with a single passenger in each and that the increase in singleton coincidences is even less likely to be by chance. When I shift the selected span of days, I only find a significant increase where there is substantial overlap with the unshifted window, which suggests that there is something special about the period around an FOMC meeting. Moreover, there is a significant increase in rides within a day of the monetary-policy announcement and thus within the communications blackout. The finding of a lunchtime increase is robust to a large set of variations in modeling decisions including the definition of a coincidence.


A potential quibble with this work is that I restrict the analysis to lunchtime rides after having inspected changes over a broader set of intraday windows. I repeat the data-mining exercise but with a much larger set of windows over the period from 9:00 through 16:59, a span during which one would expect rides generally to be non-commute trips to or from workplaces. The selected windows are 10:00 through 12:59 and two days before the FOMC announcement through seven afterwards. I find slightly under 1.5 coincidences and both similar individual significance and similar significance after accounting for the data mining to those from the lunchtime analysis. The window begins before typical lunch hours, but roughly four-fifths of the increase occurs between 11:00 and 13:00. Estimation of the increase over 9:00 through 16:59 yields the same span of days and an increase of slightly over 1.5 coincidences, so the increase is both concentrated in a span containing typical lunch breaks and does not reflect a shift of rides from one part of the longer span of workday hours to another. I obtain broadly similar results when I again only consider singletons.


Despite the strong evidence in support of an increase in lunchtime meetings, little can be concluded about who is traveling from the New York Fed or what their roles at or their connections to the Fed might be. The regular observation of potential return taxi trips would support identification with individuals based at the main New York Fed premises, but I generally do not see taxi trips to the New York Fed with pick-ups after and close to the coincidental lunchtime drop-offs. It is possible that taxis are taken when punctuality is more important, and more economical options like public transportation are taken on the way back. Alternatively, and in light of travel time, meetings might be stops on the way to other locations. 


Interestingly, I find that direct rides and coincidental drop-offs are particularly elevated around monetary-policy meetings in 2012, the year of the announcement of the third round of quantitative easingQE3—and the introduction of an inflation target. The increases in post-blackout direct rides and in lunchtime coincidental drop-offs within a day of the FOMC announcement appear to be largely driven by that year. The counts of post-blackout direct rides are especially high for the meetings immediately before and after that which approved QE3. In light of contemporaneous developments, information-seeking or clarification would not be implausible. One might hypothesize that information-seeking would have increased in 2013 around the so-called taper tantrum, but staff might have become more cautious with communication in the wake of a late-2012 leak of particularly sensitive FOMC information to a reporter. 


The approximately 1.6 additional meetings around a FOMC announcement inferred in this working paper might seem like a small increase, but a low absolute value should not be confused with insignificance. Were the taxi data to reflect the entire variation in interactions between the Fed and the commercial banks, it would not follow that that the extra opportunities for information flow would necessarily be of completely trivial importance for the banks and for information flow into markets. Even the occasional insight into non-public Federal Reserve data and discussions provided accidentally by an unwitting New York Fed insider could be highly profitable, and accidents of history like insiders’ social ties or close working relationships could give some firms an advantage over competitors around FOMC meetings. 


Taxis are, of course, not the only means by which individuals can interact; the commercial banks are only a subset of New York City’s financial institutions; New York City is not the only city where Federal Reserve insiders have access to sensitive information, and I have not even fully mined the taxi data. The large increases in percentage terms and their high statistical significance suggest a broader increase in commercial financial institutions’ interactions with the Federal Reserve.


At the same time, one must be cautious in how one interprets the inferred increase in meetings. Even the full set of extra interactions around FOMC meetings might involve a fairly small number of individuals who are not representative of the Federal Reserve System broadly, and the taxi data can at most provide evidence of an increase in opportunities for information flow. The uptick need not reflect a mutual desire to communicate about monetary-policy matters beyond what has been widely disseminated. Meetings both during and after the blackout could be purely social engagements or discreet Fed information-gathering regarding bond-market conditions pertinent to the implementation of monetary policy, but valuable information could still flow accidentally. Even a non-verbal reaction to an interlocutor’s statement can provide a valuable signal.


Additionally, Federal Reserve staff who broadly restrict their interactions with outside parties during the blackout could address pent-up demand by scheduling an above-average volume of meetings in the days after its end. While the blackout may reduce the risk of leakage of preparations for FOMC meetings, it may consequently also increase opportunities for outside parties to gain insights into their proceedings and support material over the following days. 


The micro-level behavior that can be gleaned from mobility data also has implications outside of academia. With the growth of ridesharing firms, it is timely to consider how these data might be used and what their value might be. As demonstrated here and elsewhere, the absence of explicit identifiers does not guarantee anonymity, and mobility data raise privacy concerns. Even if institutions and individuals are extremely cautious in guarding their own movements and in controlling the usage of data on themselves, insights into their activities can still be obtained from the movements of those with whom they interact. Further studies on what can be inferred from mobility data can yield a social dividend in the identification of data features that should be treated as personally identifiable information, in the development of stronger anonymization and in guidance on how individuals and firms can guard themselves against the exploitation of their movements. 




Cieślak, Anna, Adair Morse and Annette Vissing-Jørgensen. 2016. Stock Returns over the FOMC Cycle. Available here.


For more on this subject, listen to the “Strange Bedfellows” episode of the Capitalisn’t podcast: 



Disclaimer: The ProMarket blog is dedicated to discussing how competition tends to be subverted by special interests. The posts represent the opinions of their writers, not necessarily those of the University of Chicago, the Booth School of Business, or its faculty. For more information, please visit ProMarket Blog Policy.