Innovations in Data in Economics

Imagine you are tasked with investigating the effect that household income changes have on a certain variable, such as the risk of war. But unfortunately, war can affect growth, so how can you disentangle the twoway causality? Check the weather.

In fact, the above approach is precisely the approach used in one of the most influential papers on the relationship between economic growth and civil violence. As Collier notes in The Bottom Billion, because many developing countries depend on agriculture, getting too little or too much rain can severely affect growth. But fortunately for us economists, "prospective rebels do not say, 'it's raining, let's call off the rebellion'". As such, rain functions as an instrumental variable that allows us to proxy for the effect of growth on war, but avoids the effect that war has on growth. Besides in the study of civil conflict, rainfall shocks have long been used to investigate a diverse range of issues, which can range from the role of remittances as insurance, human capital accumulation, and sex-selection. While rainfall shocks seem like quite an obvious tool after the fact, I cannot help but smile at the thought of using them as such a, pardon the pun, instrumental part of research on development.

It also makes me smile because it excites me about what other data sources economists will have to leverage in the future. For example, an important part of Mian and Sufi's work on the effects of subprime mortgages was Saiz' house price elasticity data. Saiz calculated house price elasticities in metropolitan areas based on very specific geographic properties such as the percentage of area covered by water or the presence of steep terrain. He was able to generate such a thorough dataset by using satellite and topological data. Such computations, while impossible a few decades ago, are now much simpler. From the comfort of my apartment, I can easily pull up a street level map of New Delhi* and customize it using open source R. And if even I can manipulate such powerful tools from the comfort of my laptop, just imagine the new opportunities that could open up as the result of concerted research.

Other writers have commented on this "new generation" of economic data, but I think the studies discussed above add a little color on what more data really provides us.

It's tempting to say that more data will give us more correlations to work with and better predictive power. This is not necessarily the case as the number of spurious and uninformative correlations necessarily increase as the amount of data analyzed rises. However, something Big Data does give us is a better way to organize all the "natural" data sitting out there in the world. When Watson was introduced, attention shifted to the possibilities that a "personal Watson" could have on tasks that involved large database searches, such as medical care or legal research. There is no reason for economists to not share in these benefits. Many clever studies pivot on a very clever design, whether rainfall shocks or regression discontinuities because of geography. Thus Big Data may become less of a tool for direct prediction, and instead become an indispensable tool for economists to identify and deploy increasingly clever instruments and natural experiment designs.

This kind of "data mining" would not be so much as for finding correlations but to enrich the datasets that we have available. As I found out this year working on a housing finance project working with the AHS, privacy is a big deal in surveys. But with the possiblity of estimating non-economic public variables such as weather or geography, we have ever more powerful tools for estimating parameters for large groups while preserving the privacy of individual people. And even if merging individual entries is always difficult when comparing multiple datasets, such common public variables would allow us to create a base set of variables to enrich any dataset and analysis.

This change in data capabilities also has implications for the intellectual tools needed by economists to understand the data. While rigorous econometrics, especially spatial econometrics, will stay very important, it may become more important than ever to have a solid foundation in economic history. In the wake of the financial crisis, it has been fashionable to talk about how economic history would have given us a better idea of how to respond to the crash. Yet even beyond these policy implications, a better understanding of economic history could motivate the mining of old data sources, such as newspapers.

A Google scholar search for "rainfall shocks" or "rainfall shock" yields about 1400 results. What will be the future analogous tool for other fields of economics?

*On google maps, go to New Delhi and start scrolling to the west. While you are amazed by the ability to discern individual streets and apartment buildings, observe the dramatic change to a checkerboard of individual farm plots. In fact, the first time I saw this I thought the graphics resolution on my computer messed up the rendering.

A Beginner's Observations with STATA

This year, I have been occupied with a real estate finance project through the Undergraduate Research Opportunities Program here at the University of Michigan. Below are some observations about STATA that may be of use for some people.


1. Macros are useful, but rather peculiar. As a first time user, I found it very difficult to differentiate between when I should use "`variable'" and `variable'. From what I see, any time that the variable is meant to denote a string, it should be enclosed as "`variable'", whereas if it supposed to be a variable call or something of that sort, it should be`variable'

2. STATA is programmable, but not necessarily a 'programming' language. When working in STATA, you don't really have the flexibility that you would have in R or Python. It's difficult to make arbitrary functions, and sometimes I just want to say "int x = 4;", but there's no natural way to implement that. Another STATA construct that I had to familiarize myself with was the "foreach" command. It does do a natural way of iterating over an array in other languages, but I often get caught typing "for" by mistake.

3. If only I understood if's, my life would be easier. In STATA, there's two kinds of if's, one is a qualifier, and one is the standard if in programming that changes the flow of the program. For example, if I were to say

count if smsa == 0320

It would count the number of observations whose smsa variable entry was 0320. On the other hand, if I had

if(`nat'){
di "Merging with smsa control"
merge 1:1 smsa control using "tmortg`y'_`d'"
}
else{
di "Merging with control"
merge 1:1 control using "tmortg`y'_`d'"
}

The program would merge the file by smsa control if `nat' is true, and it would merge by just control otherwise.

4. On the above note, it may be useful to put a few display statements in your code. It's helpful to see  in the log where the program went, especially when there's these flow control issues.

5. While writing functions may be hard, it is certainly not impossible. To this end, the guide by Roy Mill was invaluable for me. I was able to more effectively abstract my code and reconcile it with my programming instincts.

6. Careful with do files and local variables. After I declared local variables in my do files, I could not access them once I was back in interactive mode. However, I had no problem with those local variables when I was working within the do file.

And for those STATA veterans among you, is there any way to do error handling? In Java, C++, Python, and R, there are tryCatch constructs that can keep the program running even if a variable is missing. This would be useful because it would allow the program to try to do something, and then if that doesn't work I would like to make the program go down a different path.

Any additional advice on STATA would be appreciated. Hope my observations can help some others avoid (too much) frustration.

Reaching for Yield

Reaching for yield from Yichuan Wang

This was a presentation on 'reaching for yield' that  I prepared for my investment club, Michigan Interactive Investments. The argument is the same as in Jeremy Stein's speech. In low interest rate environments, pension fund managers go into riskier assets in order to reach their benchmark yields. Therefore, this should result in high beta stocks having lower risk-adjusted returns. I test this hypothesis using stock price data from the past five years on almost all stocks listed on the U.S. exchanges, and find that there did seem to be a substantial amount of 'reaching for yield' in 2011 and 2012, and that in 2011 this phenomenon was concentrated in the large cap stocks whereas now more of the effect seems to be with the small cap stocks.

While I'm not that confident on the exact quantitative magnitudes, reaching for yield does seem like a new channel through which nominal shocks can have real effects. Given the plausible assumption that most benchmarks are nominal in nature, then deflation has the added cost of inducing excess risk taking. This suggests that after financial crises, monetary policy should be even more aggressive in order to minimize the extent of the reach for yield.

I do think this investigation brings up an important methodological issue. I believe future macroeconomic research will rely much more heavily on observations from financial markets. This was the method recently used to measure the effect of sticky prices, and this technique of disaggregating financial data to lend support to macroeconomic theories is quite intriguing. The advantage that this has over traditional data sources is that you have a much higher frequency data stream in finance. This allows more in-depth analysis on focused time periods -- something that is much harder to do with regular CPI data.

In addition, through working on this I have found data analysis through open source R to be quite powerful. I generated all the plots in the presentation with the R package ggplot2, and all the stock price data came from Yahoo Finance interfaced through quantmod. Coupled with the powerful regression algorithms in R, I could generate the desired coefficients from weighted regressions and draw them on a plot.

What Would Stein Do? - Monetary Policy and Financial Regulation

Should monetary policy play a role in financial regulation? Although Federal Reserve Board Governor Jeremy Stein argued in a recent speech that it does, monetary policy wonks such as Scott Sumner have not taken Jeremy's suggestions well. While I am sympathetic to both narratives, I feel that they miss the boat on what monetary policy as financial stability regulation would look like. When reading supporters such as MCK or opponents such as Ryan Avent, Matt Yglesias, and Scott Sumner, I find that the debate is focused on the short term interest rate response to financial risks. However, the focus on only the short term interest rate and its relationship with financial regulation is incredibly narrow and ignores important research on alternative monetary policy tools.

It may be useful to review the positions of the involved parties. According to M.C.K, monetary policy needs to be conducted with an eye to financial stability for two key reasons. First, extended periods of low interest rates encourage excessive lending. This causes bubbles to grow and results in large macroeconomic effects once they pop. Therefore monetary policy should be conducted to pop bubbles before they become major economic threats. Second, monetary policy is more effective than specific microprudential regulations such as inspecting bank balance sheets because financial innovation can allow banks to hide much of their conduct. Because all firms face the same set of interest rates, monetary policy has a better chance of "getting in the cracks" of financial markets and better protecting the system. Thus because monetary policy has wide reaching effects, it should be an important part of any financial regulation toolkit.

On the other side, some more Monetarist authors such as Scott Sumner or Matt Yglesias argue that using the short term interest rate as a regulatory tool is misguided because what matters most are macroeconomic variables such as nominal GDP or employment. Because monetary policy is such a blunt instrument, raising interest rates to pop a financial bubble is akin to fumigating an entire house to get rid of one patch of mold. The instrument is out of proportion with the threat. If the central bank stabilizes nominal GDP  the impact of financial crises on aggregate demand should be minimal. Moreover, it's not even certain if central banks can accurately identify bubbles. Even though the U.S. housing bubble seems obvious now, it's not exactly clear that policy makers could have identified it in real time. Therefore using monetary policy to pop bubbles is unlikely to be very effective and would result in substantial collateral damage.

Were the Fed to use monetary policy as a financial regulatory tool, the dilemma would be that one instrument, the short term interest rate, cannot simultaneously stabilize nominal GDP and the financial sector. One cannot address two problems with one tool. Fortunately, Stein has been working on an alternative policy regime to solve this problem. Once his alternative is considered, we can see how monetary policy and financial regulation can more effectively be integrated.

Stein's core proposal is that the central bank can be both a financial regulator through a combination of strict reserve requirements and paying interest on reserves. In Stein's model, the central bank can guide nominal variables through a kind of Taylor Rule for the short term interest rate while affecting the financial sector by manipulating the spread between interest on reserves and the short term interest rate.

To understand the logic behind Stein's proposal, we need to understand why the financial sector can be so unstable. Stein argues an important cause is an excess of short term debt. In particular, excessive short term debt raises the probability of fire sales, and the risk of a fire sale creates social costs that banks cannot internalize. Because reserves are required for short term debt issuance, a tax on reserves helps to constrain short term debt creation. In this way, a reserve tax helps internalize the costs of certain financial frictions, and therefore is an important tool for macroprudential regulation.

Stein proposes to implement such a reserves tax by first subjecting short term debt to stringent reserve requirements and then by varying the gap between the short term interest rate and the interest rate paid on reserves. To see why the gap between the short term interest rate and the interest on reserves is a tax on short term debt, recall the relationship between reserves and debt issuance. In a world of reserve requirements, the bank must hold a certain amount of reserves in order to issue the new debt.  Had the bank been able to lend those reserves out, they would have earned interest equal to the short term rate. This is the gross tax on reserves. But because the bank is also compensated with interest on reserves, the net reserves tax (heretofore known as the 'reserves tax') is the difference between the short term rate and the interest on reserves.

Under this framework, although both the short term rate and the interest on reserves are nominal variables, the spread between them is a real variable that serves to penalize issuance of short term debt. The pre-crisis policy of not paying interest on reserves meant that the short term rate was the entirety of the reserve tax. Therefore any desired increase in the reserve tax required the Fed to raise short term rates one-to-one. However, if the central bank were to pay interest on reserves, it can hold the short term rate constant while increasing the reserve tax.

In other words, policymakers can regulate debt maturities without deviating from existing interest rate rules. By increasing the gap between the interest paid on reserves and the short term interest rate, the central bank can penalize the issuance of short term debt without having to raise short term rates. In doing so, the central bank can try to stabilize nominal GDP while reducing short term debt fragilities. The central bank can engage in macro-prudential regulation while also staying faithful to its price stability and employment objectives.

No doubt, there are some problems with such an approach. The most obvious one is the case in which the required reserves tax is higher than the interest rate required to maintain price stability. In this case, since the interest rate cannot be lower than the rate on reserves, satisfying one mandate would necessarily ignore the other. To avoid this problem, the central bank could raise reserve requirements on short term debt. The intuition behind this is that for each additional dollar of short term debt, the bank would need more reserves. As a result, the effect of the reserves tax would be magnified, thereby lowering the optimal reserves tax below the necessary short term rate.

Another concern is that raising the reserve requirement could substantially disrupt the functioning of banks. Because banks have become used to certain reserve ratios, they would have a hard time adapting to higher reserve requirements, thereby reducing the money supply and causing more uncertainty. I see three ways central banks could mitigate this. First, they could announce the change in reserve requirements ahead of time and allow banks to more smoothly transition into the new regime. Second, the reserve requirement could be raised when there is a large supply of excess reserves, such as right now. Third, if nominal GDP slows down substantially, the Fed could engage in other "unconventional" actions such as Quantitative Easing to inject assets into the system. Moreover, raising reserve requirements as a part of deploying a new macroprudential framework may make certain Basel III requirements obsolete. In particular, the Basel III proposal of defining a stable funding ratio to limit short term debt issuance may become unnecessary when the Fed can control such a ratio with much more precision through a reserves tax. By making other reserves regulation obsolete, a new policy regime with interest on reserves could actually ease the regulatory burden on banks.

Neil Irwin, in his post on Stein, likened raising interest rates and popping bubbles to fumigating an entire house to get rid of one patch of mold. However, Stein's policy proposal would not be so dramatic. It would be more like a dehumidifier, addressing the risks of mold while still allowing people to live in the house.

Thus the debate over whether the short term rate should be used to moderate credit cycles seems a bit silly in the context of what Stein's papers have been suggesting. It is almost as if in all the arguments over what Stein said, people have forgotten to look at what he has written in the past. The alternative approach with interest on reserves and stricter reserve requirements can coexist with the price-employment mandate. In addition, such a strategy provides monetary authorities more tools to interact with the very large financial sector. In this way, the effective integration of financial regulation and monetary policy should improve macroeconomic stability, not worsen it.

Market Monetarism and Finance

As market monetarism starts to become more mainstream, I have started to take some time to think about what yet has to be done to develop this new brand of monetary theory. One issue that recurs in my thoughts is that market monetarism needs to help develop a richer understanding of financial dynamics.

One of the strongest justifications is that a richer understanding of financial linkages would help untangle the dynamics of monetary policy under different regimes. Scott Sumner argues that monetary policy works not with long and variable lags, but rather long and variable leads. Because agents are forward looking, expectations of future nominal GDP significantly affect current economic activity. The strongest evidence for this comes from the financial markets. For the United States, Marcus Nunes has done quite a bit of work charting the immediate effects of monetary policy hints on inflation expectations:

We also see similar evidence in the international arena, whether Japanese, Swiss, Hong Kong, or American.

However, the chart is incomplete. Past studies do suggest that the effect of interest rate cuts are not felt until several months after the initial policy declaration. While there may be identification issues with those studies, they do open up the possibility that monetary policy does not act as quickly as market monetarists would hope. In this context, a hybrid approach may be more accurate. While monetary policy leads the financial sector, it is likely that monetary policy lags in other "real" sectors, such as manufacturing.

This synthesis of both monetary and financial dynamics is especially important given Lars Christensen's argument that "there is probably no better indicator for the monetary policy stance than market prices." We know from the financial literature that certain phenomena, such as excess volatility, seem to defy the typical market monetarist use of the efficient market and rational expectations hypotheses. This is not to say policy would be better guided by the arbitrary decisions of central bankers, but rather that a move to market based signals needs to be grounded on better a theoretical and empirical understanding of how monetary policy and financial signals lead other parts the real economy.

As an example of this, we can take a look at the relationship between TIPS inflation expectations and PCE inflation. For those who don't know, the TIPS spread is the interest rate differential between the 5 year inflation protected treasury and the regular 5 year treasury, and therefore is a measure of what investors expect inflation to be over a five year time horizon.

Under the rational expectations hypothesis, expected future inflation should be a reasonable estimate of actual future inflation. By the efficient market hypothesis, these expectations should then be expressed in the 5 year TIPS spread. However, for the years during which we actually have data on how the 5 year TIPS compared against the actual inflation rate, performance is quite poor:

This evidence suggests that even market forecasts can be unreliable. While they can sometimes be a good indicator of future performance, in other times they can be unacceptably wrong. In the above example, the relationship between the TIPS forecast and actual inflation was so wrong that it was negative. While such data points may be washed out in the long run, the 5 years of flawed predictive capacity that it would have given should give any policy maker pause.

However, the TIPS spread is actually quite a good predictor of contemporaneous inflation. Below I plotted each month's PCE inflation rate with that month's average TIPS spread, and find that the linear prediction (red points) does a good job of measuring current month inflation:

This suggests that while we may not be able to use market signals to predict with precision, market signals do carry significant information content. Instead of waiting for each month's CPI report with bated breath, we could simply consider the financial data that is always available to us. This resembles my conclusion from looking at forecaster data. Given that we have reasonably accurate forecast, monetary policy should target those forecasts. When bad forecasts come in, central bankers can signal that they are ready to ease monetary conditions if the bad conditions materialize themselves. The trick here is to make sure that the information hiding in market prices can make its way into policy, and a better understanding of the relationship between finance and macro is an important step in that direction.

The Unexpected Implications of Expectations

As Market Monetarists, we always stress that expectations matter. But how can we test this hypothesis? One way we do this is to use financial evidence such as TIPS spreads to show how changes in monetary policy expectations directly affect market conditions. Evan Soltas recently tried to use a different method: surveys. He put together a series of graphs documenting changes in forecaster expectations around the time of the financial crisis, and claims that the graphs suggest that a "sudden collapse of short-to-medium expectations...could be more important than current-quarter NGDP."

But I disagree with Evan on how we should interpret such results. I took a look at the Survey of Professional forecaster data and restrict my analysis to the Great Moderation period since 1990. I also focus in on the 1 year forward NGDP forecast, as the 1 quarter forecasts give qualitatively similar results.

Like Evan, I too observe that current nominal GDP expectations are related to real variables, such as unemployment. However, the overall relationship is quite weak. NGDP expectations can only explain about 5% of the variation in unemployment. Moreover, the slope estimate is likely to be biased downwards as autocorrelation means the true slope is even closer to zero.

However, even if the correlation were stronger, it still would not say anything about the causal effect of future expectations on current conditions. Any observed correlation could simply be rational expectations at work, with causation running from unemployment to NGDP. Because unemployment is high now, it would be rational to assume that future nominal GDP will be slightly lower. Even if the Fed were more powerful, there would still be some imperfection in the implementation of monetary policy that would cause expectations to shift downwards.

To control for this, we need to look not at the change in expectations, but rather changes in surprises. A big theme in nominal GDP disucssions is that nominal prices are sticky. Therefore, when NGDP falls below trend, because past nominal contracts were set under the expectation of the higher trend, markets fail to clear and we have a fall in real growth. Therefore, if this transmission channel were true, when NGDP falls below what was expected, we should expect to see a significant impact on unemployment.

In other words, we should consider whether actual nominal GDP hit forecasts or if it fell short. This way we can construct an error index that measures to what extent forecasters over or underestimated. Positive numbers denote when actual nominal GDP outperformed the forecast, and thus the dramatic fall into negative territory during the financial crisis reflects the unexpected nature of the nominal GDP shock.

While it certainly did fall during the financial crisis, if you use ordinary least squares regression on this index against variables of interest, such as unemployment, you will not find any kind of systematic correlation. However, if you consider only the extreme cases in which the forecast undershot reality by more than 4%, then you do get a significant negative correlation:

Perhaps this evidence suggests that expectations have a nonlinear impact on unemployment, but at that point we are drawing epicycles that the regression evidence does not warrant.

Does this all mean that expectations are useless? On the contrary. When investigating these expectation surveys, I did manage to uncover the following chart comparing forecasts and actual nominal GDP growth. I lagged the actual NGDP by 1 year, so it is easier to compare how the forecast compares with actual growth.

What we can see is that the forecasters, while not perfect, still do a rather good job of identifying times of distress. Given that forecasts do carry information, then this opens up a role for policy to lean against the wind. The Fed, instead of waiting for all the data to come in, could use a joint forecast-contemporary data criterion. If the forecasters are projecting slow future growth, then the Fed could announce that it is aware of a potential problem and prepare the necessary policy machinery, conventional or otherwise, to combat that threat.

Expectations matter, but we need to be clear on why. While they may have a direct impact on growth, expectations also serve as a crucial lens into the future and can carry information content for policy makers. Armed with such tools, monetary policy can turn towards the future, lean against the wind, and in doing so remedy demand shocks before they start to hurt. 

Forward Looking Markets in Japan

Even though Shinzo Abe has not yet cemented in a new era of Japanese monetary policy, markets have already been preparing for such an event. The market response is a good illustration of how markets act in a forward looking manner.

First, we need to identify when the international community started to focus in on Shinzo Abe. For this, we can look at Google's international search intensities:

From here, we can identify the time around December 15th to be the starting point. Now let's take a look at how markets have responded since then.

First, the exchange rate. Easier monetary policy raises aggregate demand, accelerating NGDP growth and depreciates the currency. Even though the statutory monetary policy framework hasn't changed, the yen has already lost around 7% of its value relative to the dollar since December 14th:



Exhibit 2, the Stock Market. In the same time period, the Nikkei 225 index has risen over 9%. To give this some perspective, almost one-third of the gains over the course of the past year have been the result of the past month of advances.



These statistics are quite striking when you think about the policy controversy surrounding Japan's "lost decade" and the hundreds of trillions of yen that went into public works programs.

Also, these statistics are quite interesting for the "long and variable leads" perspective on monetary policy. Historically, the money supply has not had a large effect on real variables such as GDP. In fact, if you really dug into the 5 year rolling correlations, you would find that, most of the time, quarterly money supply growth actually had a negative correlation with quarterly real GDP growth in that period.



Yet just in the past month, a few "open mouth operations" have seemed to dramatically change market perceptions of monetary policy. This suggests that the mechanical money printing is quite powerless without effective expectations management. As such, there should be significant gains to Shinzo Abe's drive to change the Japanese monetary regime.

On a separate non-economic, purely speculatory note, I'm not sure how the nationalism issues brought up by Noah Smith would interact with this drive to make monetary policy more inflationary. On one hand, if monetary policy is seen as a driver to achieve national ambitions, then the nationalist drive may reinforce the increase in aggregate demand. That would be good. But the worrisome possibility, which worries me more than the possibility that Shinzo Abe will give up on monetary policy, would be that the monetary policy stimulus will be too effective and further stoke nationalist ambitions. The last thing we need is some stupid quibble over the Diaoyu islands that leads to a face-losing and economy destroying outcome for everybody.

Update 1/19/13:

Scott Sumner emails me with the following comment:

"Abe actually started pushing for a 2% inflation target during the campaign, in mid-November. The day of his first speech is the exact the the huge stock rally began, and the exact day the yen began falling sharply. I did a post that day, or perhaps the next. 

http://www.themoneyillusion.com/?p=17736 

It's hard to disentangle money printing and expectations. Think of the following thought experiment: The central bank doubles the money supply and is expected to cut it in half 3 days later. Everyone agrees that there is little effect on markets or AD. So in some sense we are always implicitly making assumptions about monetary policy. On the other hand, I can easily envision where big changes in the money supply also contain information about future expected monetary policy, even if not made explicit. So I can envision QE "working" at least to some extent, even w/o an explicit promise regarding future policy."

Looking back at the data, I should have been more careful with the timing issue. Although the November 15th spike looks minimal, it is only because the December spike was much larger. Focusing in on that one month of the Google data yields the following graph:




On the relationship between money printing and expectations management, I agree with Scott that money "printing" policies, such as QE, can have significant effects even without explicit open mouth operations. The Fed doesn't need to announce a NGDP target for those policies to have effect. But the point with Japan is that effective communications policy can enhance monetary policy. Given that money supply expansion hasn't been sufficient in the past, perhaps a more powerful target will do the trick.