Financial Stability Discussion in Fed Communications
This document analyzes how financial stability is discussed by Fed officials during the 1970-2023 period. It relies on both public speeches and FOMC transcripts. It first gives an overview of the salience of this issue in all documents, before focusing only on FOMC transcripts, and more specifically on FOMC interventions that also mention monetary policy issues.
1. Financial stability
In a first step, I looked at both public and private speeches - if we have them, why not use it to see the relevance of such issues for all policymakers. That said, I also do focus more on FOMC transcripts below, but always with only financial stability keywords. The rationale is that any discussion of financial stability in FOMC meetings is in itself interesting for our research question, as these meetings are there to discuss monetary policy decisions.
In section 2, I will filter more radically, looking only at FOMC meetings interventions that mention financial stability AND monetary policy keywords. But bear with me, I think there are interesting insights in this first step already.
1.1. “Financial stability” vs “monetary policy”
I decided to first plot the share of “monetary policy” and “financial stability” in the total amount of words in both public speeches and FOMC interventions to gauge a bit the evolution of its salience through time.
This already reveals several key patterns:
- Financial stability has lower salience than monetary policy
- Financial stability is mainly discussed publicly in speeches until 2010
- Public communications seems quite strategic: high salience in the asian crisis, in 2001 ; in 2007/2008, but each time no actual discussion internally
- Private discussions in FOMC meetings rise after 2012
1.2. Financial stability more in depth
I then create a more complete dictionary with the following keywords:
‘financial stability’, ‘financial instability’, ‘systemic risk’, ‘systemic importance’, ‘punch bowl’, ‘financial vulnerability’, ‘financial stress’, ‘financial risk’, ‘financial cycle’, ‘financial imbalance’, ‘financial excess’, ‘asset bubble’, ‘credit bubble’, ‘leverage cycle’, ‘leverage ratio’, ‘financial crisis’, ‘market stress’, ‘market turmoil’, ‘financial distress’, ‘bank run’, ‘fire sale’, ‘contagion’, ‘interconnectedness’
EDIT: following Matthias input, I also add: ‘credit crunch’, ‘exuberance’, ‘elevated asset prices’, ‘leaning against the wind’, ‘lean against the wind’
I then plot below the salience of these keywords along four dimensions: Public / private (panel A), Position (Panel B), Gender (Panel C), Speaker (Panel D). To keep things readable, I only include the top 10 speakers on the issue.
Panels B, C, D focus only on internal FOMC meetings, as this is what is of interest to us.
There is a big big hit in 1990 which is caused by “credit crunch”. I think we should exclude it?
Otherwise, not much new information here: - The public/private divide is confirmed even with a more complete dictionary - Position or Gender do not seem to play a big role - but a few Speakers stand out
1.3. Exploring deeper the speaker level
To build on this last insight, I look at all speakers and plot their private vs public salience vis à vis financial stability issues, again with the same dictionary. Green lines show salience in internal discussions, blue line in public speeches.
Pay attention to the fact that the scales are not the same for readability reasons! Fed Board Members can hence go super high in their financial stability salience. Also keep in mind those are relative weights, so speakers that are not very talkative but do often favor those types of discussions are favored over speakers that are very vocal on all topics.
With that in mind, I made the same plot below with the total number of mentions of financial stability in private / public interventions.
Tarullo, Bernanke, Rosengren, Dudley, Bullard, Kocherlakota and Evans appear as relevant players. George is not that high up.
Notice how certain players (e.g. Tarullo) speak a lot in public speeches about this issue compared to others (e.g. Kocherlakota)
1.4. Any institutional / structural drivers?
Then, I wanted to see whether or not Matthias’ hypothesis on Fed Boston / Fed NY being structurally more inclined to deal with these issues, given insurers for the former and monetary policy implementation for the second.
In relative terms, NY actually comes just after Minneapolis, but they both rank in the top 3 in private interventions (in green), and are the two most vocal ones in their public speeches (in blue)
But is this really a “regional fed” effect, for institutional or strucutral reasons? Or could it be simply a “speaker effect” that happens to be in those two institutions? To put things differently, do Rosengren speaks about financial stability because of his position as the Fed Boston president, or is it because of idiosyncratic features of his persona that distinguishes him from predecessors?
To look at this, I figured that looking at all the successive president for each central bank may help. And I was not disappointed:
Pretty amazing plot for Rosengren (as early as 2006!?) and Dudley (he joins in 2009, so easier I guess). Keep in mind this is aggregating all types of communication, both public and private, but I guess the structural/institutional rationale kind of takes a hit. The ones before and after those two guys don’t give much attention to the issue.
And what about the others? I did a second plot that looked at all 12 Regional Feds, because why not. This time, I also separated the public vs private communication.
I think this is my favorite plot. It is very dense in information, so do zoom in. You can see both the impact of new nominations, as well as the public vs private dichotomy (full vs dotted lines), that can sometimes be massive. Maybe because there just no institutionalized moment in which to bring these issues to the discussion (before 2012 at least) ; or maybe because these guys are all show and no real action.
What I also find fascinating is the hysteresis effect: all guys that started to talk about this when taking office just after the GFC kept talking about it throughout their mandate. When they were replaced, it seems like the issue died. San Francsico and Minneapolis would be the only exceptions.
On Minneapolis, it’s amazing to see Kocherlakota is actually more vocal internally first before trying to take the issue publicly. Maybe something to dig there?
Atlanta is also super interesting about the (likely credit crunch linked) discussions in 1990.
1.5. How do one talks about financial stability?
This is only the very beginning of my work on this, so take these preliminary results with a grain of salt. What I wanted to do next was trying to see how these guys speak about financial stability in their speeches.
What I did is simple: keep only sentences including at least one financial stability keyword, and look what other keywords usually appear in the same sentence.
My goal being to find “camps” among speakers, I first did a simple wordcloud based on these financial stability sentences, in which most frequent 100 words are plotted with a size proportional to their frequency. Maybe they don’t use the same words and we can separate them this way.
Of course, this is not done on the raw text: it requires in pre-processing steps to remove words that are too frequent (“stopwords”), make sure everything is lowercase, remove punctuation, numbers, and even “stemmatize” the words to make sure that “banks”, “banking”, “bank” are not counted in a separate fashion, but are indeed counted together under their root “bank”.1
Well, as you can see, the problem is that the very common (policy) words are just driving most of these wordclouds, making it quite unlikely to grasp meaningful insights.
With that in mind, the next (and last) thing I did was trying to compute the most distinctive words for each speaker, namely words that appear often in their interventions. but not in other members’ interventions. To have as many observations as possible, these next plots look at all sentences that mention financial stability concerns, whether in private FOMC meetings or in public speeches.
This is already pretty cool - see e.g. the importance of the Mexican crisis or of the Norwegian experience for certain speakers. I also love the climate insight for Brainard.
The same, but by decade, also shows interesting patterns:
That said, keep it mind this is also incorporating the public speeches by these actors, which means these may not really be about monetary policy at all.
2. Monetary policy FOMC interventions only
Let’s try to restrict the analysis to the FOMC interventions that explicitely mention monetary policy issues.
First, the dictionary:
“monetary policy”, “interest rate”, “federal funds rate”, “policy rate”, “balance sheet”, “inflation”, “deflation”, “disinflation”, “quantitative easing”, “QE”, “large scale asset purchases”, “forward guidance”, “open market operations”, “reserve requirements”, “discount window”, “policy accommodation”, “policy stance”, “risk-taking channel”, “risk taking channel”, “financial frictions”, “credit channel”, “bank lending”, “monetary transmission”, “price stability”, “policy communication”, “policy guidance”, “policy path”, “policy normalization”, “lift-off”, “tightening”, “easing”
Only 1,162 FOMC interventions out of 82,090 mention at least once one of these keywords AND one of the financial stability keywords above. The rest of this section investigates these excerpts.
2.1 Relevant excerpts
Here are the relevant excerpts, disagregated along several variables.
We still have the odd spike in 1990, which comes from the “credit crunch” discussions that were trendy at that time (Panel A.). Regional Fed President are above the others, but considering these guys are 12, it is the Fed Chair that talks the most about this - logical, if you think he’s the most vocal (Panel B.). Fed NY and Fed Boston are just below the Board (Panel C.), and credit needs to be given to Bullard, Dudley, Rosenberg, Tarullo, Evans and Kocherlakota for being the most vocal (Panel C.)
2.2 What keywords drive these patterns?
A long part of the discussion with Matthias was that we need to make sure our list of keyword is time-proof, especially if we want to go beyond the nineties. What if they used a different term than “financial stability” (e.g. global warming was replaced with climate change). We want to avoid anacronistic projections of today’s categories and concepts in the past to then claim this is all so new, when in reality they may have been discussing things differently.
With that in mind, here are two plots that show the relative importance of each keyword for each year
The credit crunch problem is acutely illustrated here. Of course, don’t get me wrong - maybe it is not a problem per se if it was actually a way of discussing what we are interested in.
But is it? Is it really financial stability <=> monetary policy? Or just the classic “we need to ease during a recession to avoid a debt deflation’? To me credit crunch does not mean”financial instability”, it just means banks are cleaning their balance sheets after a crash, and hence not lending a lot, which depressed activity. What we are fishing for, I guess, is more how financial risks build up before a crisis and whether and how to incorporate this uncomfortable knowledge in the monetary policy decisions?
3. Discussion and next steps
I have tried to do (i) topic modeling, (ii) tf-idf and (iii) wordfish on these excerpts, but they are simply too few, I am afraid.
The next step should, I think, be to inspect manually these interventions and judge whether or not they are indeed interesting or not to get false positives out and get a better sense of how to spot the really interesting nuggets.
On the other hand, there is also a discussion to be had on potential false negatives: could we be missing important discussions with our keywords? Especially in the begining of the period?
And then what? Manual coding?
4. Build-up vs Clean-up
After our call late december, we decided to explore a few avenues. First, to look the prominence of discussions about the build up of risk in the system during booms, VS discussions about the clean up afterwards.
4.1 Build up vs clean up dictionaries
First, I had to come up with dictionaries. The build-up dictionary contains 22 terms associated with preemptive risk monitoring, including keywords such as “bubble,” “moral hazard,” “speculation,” and “risk appetite.” The clean-up dictionary comprises 14 terms related to crisis response measures, featuring phrases like “lender of last resort,” “bailout,” and “emergency lending.”
buildup_terms <- c( “bubble”, “moral hazard”, “overvalued”, “overvaluation”, “speculation”, “speculative”, “exuberance”, “frothy”, “lean against the wind”, “punch bowl”, “risk-tak”, “risk tak”, “excessive risk”, “excess risk”, “risk appetite”, “leverage ratio”, “easy money”, “too abundant”, “financial imbalance”, “misallocation”, “asset inflation”, “minsky” )
cleanup_terms <- c( “lender of last”, “bail out”, “bailout”, “bail-out”, “bail-in”, “bail in”, “bailin”, “rescue”, “liquidity support”, “emergency lending”, “crisis management”, “clean up”, “restructuring”, “backstop” )
4.2 Build up vs clean up through time
Let’s plot the importance of these words through time. To maximize the chances of finding stuff, and considering these key expressions are quite precisely what we look for and should not lead to capture non-financial stability discussions, I search these words in ALL THE FOMC INTERVENTIONS.
Two comments: - first of all, this is not a lot of hits. Missing keywords? - second, build up terms seem to be actually more frequent
We probably lack a lot of interesting words in both dictionaries, while others may lead to too much false positives. in any case, if we do want to really look at this seriously, we’ll need to patiently spend a few days actually reading those financial-stability intensive FOMC meetings that I sent around in excel file + can be found playing around with the shinyapp, in order to curate in an iterative and patient fashion the keywords that really capture what we are fishing for there.
Besides, we may want to ask ourselves what we want to achieve with this? is this just producing an illustrative plot like we did with the migrating birds story? If so, what’s the story?
Last thing to keep in mind: we now do not look at 10 years like our previous papers, but 5 decades. The terminology probably evolved during this period. We want to avoid anachronistic behaviour, searching for present terms in the past and infering from their absence claims about the lack of interest for these matters in the past. Linked to that, you can see that not only do these terms in both dictionary have very different frequence of appearence, but this frequence evolves through time:
Quite likely that restructuring is picking up false positives, no?
Same goes for the build up dictionary:
4.3 Build up vs clean up across speakers:
Not all speakers are equal on the use of these different terms. I have several plots that all can try to show this imbalance, looking at how often each speakers uses terms from both dictionaries:
In the first, speakers are ranked in a descending order of frequency of use of all terms (on top, those that use terms from either dictionary the most often, on bottom, those that do the less frequent use of these words)
Same data, but better (IMO) plot:
Is this confirming insights you had qualitatively? Is it not? If not, let’s investigate which keywords are missing or which are driving false positive problems. If this is still not enough, could it be due to negation of certain terms (eg “there is no risk build up” or “we should not care about speculation”). Again, counting keywords can only achieve so much…
5. Specific core concepts
Then, the idea is to look at how a few core concepts are leveraged and discussed (again, through time and across speakers). We settled on exploring “lender of last resort”, “moral hazard” and “bail-in/out”.
5.1 Moral hazard
Footnotes
(Actually, we could discuss whether or not there is a rationale for not stemming words, which in some specific cases may be adapted if we think there is a meaningful distinction between policies and policy, for example - I did not see any argument for that in our use case, but keep in mind this is possible).↩︎