The
Princeton Initiative on Macro, Money, and Finance was a
fantastic event. I flew from Oakland through Houston and into Newark on
Thursday night and arrived at the Nassau Inn just after midnight. It is a nice
and conveniently located hotel. On Friday morning, September 7, I left the Inn
around 7 a.m. with my roommate, a finance student at the Haas school at
Berkeley. For us Californians, it felt like 4 a.m., but luckily plenty of
coffee was provided at breakfast in the gorgeous chemistry building on
campus.
The day began with a lecture by
Markus Brunnermeier called
"A Brief History of Macroeconomics." Interestingly, macro and finance
developed along different paths, often developing similar notions using
different languages. Now, it seems, they are converging. This lesson on history
seemed equally a lesson on the future-- a future in which finance and macro are
studied in a more unified framework and financial frictions are taken more
seriously. This conference urged us future economists to consider going in that
direction.
Professor Brunnermeier continued with a lecture on
"Liquidity concepts: amplification, persistence, and asymmetry." He
gave a convincing argument that liquidity risk is a more fundamental concern
than maturity risk. There are three types of liquidity. On the asset side,
there is technological liquidity, referring to the reversibility of investment
in physical capital, and market liquidity, referring to the specificity of
claims capital (if the second best use of the capital is still nearly as good
as the best use, market liquidity is high). On the liability side, funding
liquidity is tied to the maturity structure of debt. We saw many times in the
lecture how different classic and newer papers have incorporated one or more
types of illiquidity. For example, technological illiquidity is represented by
capital adjustment costs, or in the extreme case by a fixed capital stock.
Another interesting concept he discussed was the
"volatility paradox." An interesting implication of some of his
models is that systemic risk can build up in times of low volatility, in the
form of bubbles or imbalances. Then, once a crisis hits, this risk that has
stealthily built up in the background results in direct and indirect spillover
effects. Direct spillovers come from the direct contractual interconnectedness
of the financial sector, and have been studied a lot and found to be important,
but not enough to account for the severity of crises. Indirect spillovers, like
fire-sale externalities, credit crunches, and liquidity spirals, are as of yet
less studied, but seem much more dramatic. The thing about these effects is
that they are a "general equilibrium phenomenon": you cannot simply
detect them in the data unless you have a model (and it must be a dynamic
model) because there is not a simple story of cause and an effect. All agents
are optimizing across time and taking other agents' optimizations into account.
It was extremely helpful to get an overview of the broad
similarities and key differences between the important papers in the financial
frictions literature: Townsend 1979, Bernanke and Gertler 1989, Carlstrom and
Fuerst 1997, Kiyotaki and Moore 1997, and a number of others. I had seen most
of these papers in previous classes, but seeing how they fit together and how
the field has progressed in a logical order was enlightening.
Professor
Yuliy Sannikov gave the pre- and post- lunch
lectures on "Heterogeneous agent models with financial frictions: a
continuous time approach." The reason to study heterogeneous agent models
is due to the fact that with incomplete markets, distribution of wealth matters
(because agents cannot fully insure themselves.) Non economists may find it
strange to learn that economists usually assume that the economy can be modeled
as if there were a single representative consumer. With no financial frictions,
this is perfectly reasonable. Moreover, it comes from one of the most powerful
results in macroeconomics--if agents have access to complete markets, they will
trade securities in a way that smoothest consumption across possible future
states of the world. This then makes the math of "aggregating" agents
very nice; you can just act as if they are a single agent. I can't emphasize
enough how crucially most macroeconomic models rely on this result. This is
precisely why, if financial frictions matter, they will matter in a big way.
To present his recent work with Professor Brunnermeier (they
call it the BruSan model), Professor Sannikov began by going over a more basic
version by Basak and Cuoco (1998). A critical step in a model with
heterogeneous agents is defining a state variable that characterizes the wealth
distribution in the economy, and then finding the law of motion for that
variable. In this model, there were two types of agents, and the state variable
was just the fraction of wealth held by one of the agents. Im not sure, with
more than two types of agents, whether you would represent the wealth
distribution with a single summary statistic or if you would need to use
multiple states (maybe the number of agents minus 1?) to have high enough
information content. Since I have not taken any continuous time asset pricing
classes, I was grateful to get the basic model first, to see the notation and
basic tools for continuous time stochastic processes. A capital dividend stream
is often assumed to be a Brownian motion process, and to characterize the laws
of motion for functions of continuous variables, you can use Ito's lemma, which
is kind of like using the chain rule and product rule in "regular"
calculus. The "history" lecture in the morning mentioned that finance
began in continuous time and macro in discrete time. Since I have mostly
studied macro and barely studied finance, continuous time models are not very
familiar. I will certainly find a textbook on stochastic calculus to browse and
then take another read of BruSan.
The final lecture of the day, by Professor
David Sraer, was
"Financial frictions: empirical facts." Neoclassical models with
complete markets and no financial frictions make some basic predictions, such
as how investment should react to cash flow (in short, it shouldn't.) Taking
the neoclassical model as the "null hypothesis," can we find any
empirical evidence that would cleanly reject the null? There have been a large
number of suggestive studies but no "smoking gun." It is very hard to
conclusively show that there are financial frictions, much less what type of
financial frictions. That doesn't mean that they don't exist-- it seems,
anecdotally and theoretically and even intuitively, that they exist and matter
hugely, they are just very tricky to identify. As I mentioned, we are trying to
study general equilibrium effects. A major impediment is endogeneity. Some of
the papers that Professor Sraer discussed used difference-in-differences or
even triple differences specifications to try to alleviate the extent of the
identification. I wonder if using such specifications can ever be more than
just suggestive. What would it take to be convincing?
After the Friday lectures we had a barbecue at the Bendheim
Center for Finance. At dinner, like at the other meals, it was great fun to
meet the other students from economics departments and business schools across
the country (and even a few from schools in Europe.) I am always curious about
what the graduate student experience is like in other places, and it is also neat
to hear stories about what certain professors (whom I will leave unnamed) are like
in person. We also got to visit a bar in downtown Princeton. The downtown
Princeton street bordering campus is populated by J. Crew, Banana Republic, Ugg
Boots, and Ann Taylor shops, in lieu of the the tie dye t-shirt stands on
Telegraph by the Berkeley campus. The bar in downtown Princeton
curiously--almost eerily-- played no music. But that did facilitate more
(relatively non-econ-related) conversation.
On Saturday we heard the first presentation of the latest
BruSan paper on the redistributive impact of monetary policy. It sounds like a
simple idea but is really quite profound. In New Keynesian models, the reason
monetary policy has an effect is because of price stickiness. Monetary policy
works through its ability to alter price setting or minimize price
distortions. In BruSan's so-called "
I theory," monetary policy has an
effect through it's ability to change the distribution of wealth. The I stands
for intermediaries or for inside money. To me it seems to obvious that monetary policy affects
different people differently, and that this matters, that I could hardly
believe this was something new. I think
if you asked the average person on the street what they thought about monetary
policy, they would complain about it being unfair in some way or another, and
redistribution always seems unfair to a lot of people. But in the
representative agent paradigm, this effect does not exist, because there are
not different agents!
I am still pretty surprised that awareness of the
redistributive impact of monetary policy was not high enough to convince people
to study heterogeneous agent models more intensely long earlier. I guess it
took the crisis and associated nonconventional monetary policies to drive it
home. And adding agent heterogeneity, which we must do if we take financial
frictions seriously, is hard! The analytical tools are only in the early stages
of development. Again to give non economists an idea, at this point a
"heterogeneous agent" model may well mean you have two agents instead
of one. But that makes it way more than twice as hard. And there are so many
possible ways of introducing heterogeneous agents that it can be daunting. To
make your analysis tractable, you have to you have to make a lot of simplifying
assumptions, but need to make them carefully so that the analysis still has
some hope of being somewhat meaningful. Surely, the techniques that we today
consider pretty easy and straightforward were also considered hard and daunting
in their early stages. I am confident that the toolkit for heterogeneous agent
models will develop significantly and probably rapidly (both because it is so
highly demanded, and-- from what I saw at Princeton--so many smart people are
up for working in it.) As the toolkit progresses, the I Theory of Monetary
Policy seems one of the most natural and important applications. Moreover, it will allow us to analyze the interactions between financial stability and price stability.
The next lecturer, Professor
Ben Moll, did give us some
helpful suggestions for making life easier with heterogeneous agent models.
First and most obvious is to give agents log utility, so that consumption is a
constant fraction of wealth. Professor Sannikov also showed us earlier why log
utility implies that the Sharpe ratio is equal to the volatility of wealth.
Professor Moll also recommended using continuous time stochastic processes,
particularly when your model is to have persistent shocks, and using constant
returns to scale production functions. Finally, he told us a useful equivalence
result. You can either assume that firms own and accumulate capital, issue
debt, and face collateral constraints, or you can assume that firms rent
capital and face a rental limit. Results are equivalent, but the rental
formulation may make the model more tractable.
Professor Moll discussed "Productivity losses from
financial frictions." He made a point worth repeating, that differences in
income between countries are much larger than differences in income across the
business cycle in a given country. So arguably it is more important to study
why there are cross country income differences than to study business cycle
fluctuations. Now a lot of studies have concluded that differences in capital
between countries are not nearly enough to explain differences in income. The
main explanation is differences in total factor productivity (TFP), which
basically means differences in how effectively the economy turns inputs into
outputs. Measured TFP is the so-called Solow's residual, which is a euphemism
for "what we (economists) don't know is going on." (Maybe that's not
100% fair. Someone should correct me if not.) Residuals, by construction, are
the unexplained. To explain the unexplained, one entry point is financial
frictions. This is the path Professor Moll takes. Financial frictions can lead
to capital misallocation, i.e. putting capital to less than best uses, which
can lead to TFP losses.
How do financial frictions lead to capital misallocation? In
the model he presents, agents are entrepreneurs who are heterogenous in their
productivity and wealth. The financial friction is a collateral constraint.
They can borrow up to a certain multiple of their wealth; that multiple
represents the quality of financial markets. Depending on their productivity
and how much they can borrow, it may or may not be profitable for them to
undertake their project. It turns out that there is a productivity cutoff for
being an active entrepreneur or not. Measured TFP for the whole economy depends
on the cutoff, which depends on the quality of financial markets. My housemates
are development economists. I want to ask them how they think about TFP
differences between countries and the role of finance.
The next Saturday lectures was "Bubbles and
crashes," by Professor Brunnermeier. I learned the Brunnermeier and Abreu
model of rational bubbles in a first year class, but what didn't stand out to
me until this lecture is the following. Normally, backwards induction arguments
rule out the possibility of a rational bubble. But such backwards induction
requires common knowledge. The fact that agents are sequentially made aware of
mispricing means that they eventually have mutual knowledge of first order,
then of second order, and so on, but common knowledge is mutual knowledge of
infinite order, which doesn't happen in finite time. Game theory and information theory are taught as segments of our first year microeconomics sequence, but they are important in macroeconomic models with financial frictions.
The final Saturday lecture was "A welfare criterion for
models with distorted beliefs," by Professor
Wei Xiong. He opened with a
funny anecdote. Supposedly, economists Stiglitz and Wilson made a bet of $100
about whether a pillow was stuffed with natural down or synthetic fiber. One
believed the probability of down was 10%, the other thought 90%. To find out
who won the bet, they had to cut open (and destroy) the pillow, and they agreed
to split the $50 replacement cost. Both economists had an expected value of $55
for the bet (left as an exercise for the reader) so they both happily agreed to
it. But the bet had a negative net value! Whatever happened would result in a
transfer payment of $100 from one economist to another, and destruction of a
pillow worth $50! Bets between economists have high pedagogical and
entertainment value. Maybe other people find less destructive ways to entertain
themselves. But there are many more common trades that result from
heterogeneous beliefs and have negative net value. A social planner should be
able to make people "better off," but the question is, what beliefs
should the planner use to evaluate welfare? Xiong introduces a belief-neutral
welfare criterion. The set of reasonable beliefs is the set of convex
combinations of agents' beliefs (in the pillow example, any probability between
0% and 90% that the pillow is down is a reasonable belief.) Choice x is called
(in)efficient if it is Pareto (in)efficient evaluated using any reasonable
belief. Note that a choice may be neither efficient nor inefficient under this
definition. I really enjoyed this lecture, and wonder if I should attempt to do
research that is more decision-theoretic. I have realized that I tend to be
drawn toward topics with a common thread of uncertainty, belief formation,
information processing, and ambiguity.
On Sunday morning I woke up early enough to go running on
campus and on a nice crushed gravel waterside path. Next, the morning lecture was
given by Professor
Chris Sims on the "Fiscal Theory of the Price
Level." More aptly, the lecture was what he called a
"metafiscal" theory of the price level, talking about the model from
above and outside of it more than really going into it. This is what I love
most, to be honest. I would be a metaeconomist if that were an option, since I
spend much more time thinking about economics (in the meta sense) than thinking
about the economy, and with more passion. I am the kind of person who
likes the Introduction chapter of books best of all, especially if it goes way
into why and how they decided to write the book. Not too productive at this
stage in my career, I know. As a macro student, literally hundreds of times I have taken
this little step where you have a one period budget constraint (for the
government, say) and you sum it up over all periods (yes, back in discrete
time!) and appeal to a transversality condition to derive an intertemporal
budget constraint. The transversality condition is a restriction on asset value
as time goes to infinity(!) and all professors have a slightly different way of
explaining it to students. I understood the gist of why, economically and
theoretically, it is needed, but sometimes its implications did seem slightly
unsettling. I think Professor Sims' explanation got to the heart of why it
sometimes seemed odd. The single period constraint is an accounting identity,
but the TVC is NOT. It is an equilibrium condition. Professor Sims also argued that conventional ways of thinking about the independence of monetary policymakers and fiscal policymakers should be reconsidered.
Professors
Nobuhiro Kiyotaki and
Atif Mian gave the final lectures of the conference. Anyone who is familiar with the Kiyotaki and Moore model of credit cycles, now part of the canon for first year grad students, may be interested in Professor Kiyotaki's
newer model of banking, liquidity and bank runs. I took a
course on Empirical Macro Finance with Professor Mian last semester that greatly sharpened my interest in the intersection of macroeconomics and finance and taught me a lot about empirical identification strategies.
One thing I didn't realize before the Initiative is that
economic history is not so common to study at non-UC schools. When I mentioned
that economic history is one of my fields, people often thought I meant history
of economic thought. And though the Initiative did not have any explicit focus
on economic history, nearly every lecture brought to mind historical episodes I
have studied. An absolute must-read, in my opinion, is "
Finance Capitalism and Germany's Rise to Industrial Power" by Caroline Fohlin. To study financial frictions, it seems quite useful to study
earlier stages of financial development. It is not immediately apparent whether
less developed financial systems would have more or less severe financial
frictions. There have likely been complicated interactions between
technological and financial innovation and economic development over the
centuries. Some innovations alleviating certain frictions and exacerbating
others. My colleague
Glenda Oskar, on the job market this year, is researching
19th century capital markets. She looks in particular at a California statute
passed in 1861 that granted mining companies the right to levy assessments
(like negative dividends) on existing shareholders. She has, impressively,
collected a dataset that allows her to study when and under what types of
conditions different mining companies exercised this privilege. How did financial frictions in that era compare to today, and what can that teach us about their impact?
Closely related to economic history is the economics of
institutions, also much emphasized at Berkeley. Financial frictions depend on the "
rules of the game," which realistically are neither exogenous nor static. For now, most models with financial frictions implicitly take the institutional arrangements as given. However, both positive and normative analysis will benefit if we eventually bring institutions into the model. Many policy changes in response to crises are actually institutional changes.
The Princeton Initiative gave me a lot of interesting ideas to think about. I am grateful to everyone who made it possible.
