China's Growth: A Look Inland - Data Supplement

Today my Quartz column on the changing economic geography of China was published. In this post I intend to cover some extensions of the article that did not make the cut, and in addition go through some of my data analysis procedures so as to provide a resource for fellow students doing similar research.

A central idea is that the base unit of analysis for the Chinese economy should be the province. This is because China's massive size make its provinces as large as entire countries. For example, Guangdong, a coastal province, has 108 million residents. In comparison, Mexico only has 112 million residents and the entire Western United States only has 71 million residents. The entire continent of Europe has only around 740 million people -- a little less than half that of China's 1.3 billion. As such, lumping all the Chinese provinces together into one entity called "China" papers over so much heterogeneity in income levels and growth rates -- resulting in a very misleading picture about the actual economic situation.

To get an idea of these massive income differences and why it's important to look at provincial data, consider the stories of Guangdong and Guangxi, two neighboring provinces in southern China. In 2011, Guangdong, the relatively rich coastal manufacturing center, had per capita income of about 51,000 yuan (~$8,300 USD). Yet Guangxi, an inland province right next door, had nominal per capita income of only 25,200 yuan (~$4,100). Does it really seem plausible that Chinese growth will slow down so suddenly  that two neighboring provinces whose names differ by one Chinese character* will maintain such a large income gap into perpetuity? Given that Guangxi's per capita income increased by a factor of 3.36 from 2001 to 2011 and Guangdong's per capita income only increased by a  factor of 2.06, I would have to say no. Moreover, even if income levels do not completely converge, income growth should. Since Guangxi's income growth rate is still so high, I have to conclude that it's growth will likely be sustained for some time. Had I not analyzed the provincial data, I would have instead seen a downward trend in national real GDP growth numbers and concluded that China will suddenly slow down. But by taking into account the way growth rates evolve across provinces, I arrive at a more optimistic GDP number.

Geography is especially important given that many of the arguments made by Krugman and a recent IMF working paper center on Chinese labor markets. The argument is that since China has become richer, China has reached "peak peasant" and can no longer sustain such high levels of growth. But I'm left asking -- which provinces have hit this peak? Given that the inland provinces are still relatively poor, there still seems to be a lot of room for these provinces to grow. Although the move towards manufacturing in inland provinces may be a sign that coastal provinces are facing labor shortages, the "reach for peasants" suggests that inland China still has plenty of labor market slack left As a result, I am left quite skeptical about these dramatic bear stories for a sudden slowdown the Chinese economy.

I also want to add one more graphic to this conversation about China's growth. While the scatterplot in the column does a good job of showing convergence, I wanted another plot to just show how much individual Chinese provinces have grown in the 10 years spanning 2001 to 2011. I settled on the chart below. Besides the components in the legend, the small numbers to the left and right of each dot is the nominal per capita income (in thousands) for the specified province and year. The black number in the middle of the band is then the ratio between 2011 and 2001 levels of real GDP.

The nominal number is useful because it allows relatively quick conversions into U.S. dollars. As such, it seems that the per capita income in Shanghai is around $13,300 -- a level slightly ahead of Mexico's per capita income of $10,247 and the U.S. poverty line for a single person household of $11,344. The black multiple then emphasizes how much individual Chinese provinces have grown. These above-three multiples correspond to over 12% growth, so if a child entered elementary school in 2001, then by the time he or she goes into elementary school, GDP in that province would have doubled.

Of course, there are risks to the bull case that I present in my Quartz column.

Chief among these risks is if there's an environmental constraint prevents the inland provinces from obtaining the same levels of income as the coastal provinces. The Solow model (on which convergence is based) does not take into account natural resources, so if natural resources run out this process of convergence could fall apart. This does not have to be a hard scientific constraint either -- public outcry against environmental destruction would have a similar effect. While I agree that China does face serious environmental challenges (particularly in air and water pollution), I don't think protests will play as large of a role that people suggest. Remember that the recent large scale environmental protests -- in Zhejiang against a petrochemical plant and in Guangdong against a nuclear plant -- have taken place in the richer coast. Therefore inland China still has a way to go before this environmental constraint becomes more severe.

Others may raise the issue that the Chinese provincial data are a dangerous form of "science fiction". Indeed, it is a bit peculiar as the sum of all the provincial GDP numbers does not equal the total national GDP. But as Princeton professor Gregory Chow notes, while year to year GDP growth rates may be easy to manipulate, levels are not. Since the levels are recollected every year, measurement errors accumulate and therefore any kind of fake data becomes unsustainable. As a result, I focused on a 10 year average growth rate to resolve the issue of year to year measurement errors. Moreover, a recent San Francisco Fed economic letter found that national Chinese data seems to be accurate and consistent with a wide variety of indicators. Thus it seems doubtful that the main convergence result was just the result of data manipulation.

The bottom line is that China's great size means that attention needs to be paid to the individual provinces. On the basis of the provincial levels of growth, I am left quite optimistic about the future of Chinese growth.

If you want to try and replicate it (please do), just consult the public dropbox folder. The workflow goes from running all the STATA do files first and then transitioning into solowQz.R file to draw all the pictures. I have also included a Makefile to go through this workflow. (A Makefile executes all the code in order according to the dependencies. If you plan on doing any major work with code you really should learn a little bit on how to use them)

The one interesting methodological issue was how I used convergence to forecast future provincial growth. What I did was run a weighted least squares regression of average growth rate against initial log income, in which data was weighted by population and the estimator minimized the sum of weighted square residuals. On the basis of this regression, I assumed that the same relationship between initial income and growth continued into the next ten years and constructed measures of what growth should look like. After I had per capita income estimates, I assumed that population in each province would stay, and on this basis calculated total real GDP numbers by adding up the GDP in each province.

I had a fun time drawing the maps as well. I used R to interface with the GADM databases, and you can look at the code in chinaMap.R to get a better idea of what's going on.

If there are any more questions on code, please reach out. My email can be found on my About Me page.

*Guangxi and Guangdong literally translate to the western and eastern expanses, respectively. They are really are two sides of a lingual coin.

China's Circularity Problem

In the context of future looking monetary policy, the circularity problem refers to the problem that central banks face when they try to use market signals to guide policy. The general problem is that the market signals may include expectations of future policy in addition to their expectations of future shocks, so that the market signals fool the central bank into pursuing inappropriate policy. For example, if the private sector believes there will be a large shock to consumer demand in the future, but also believes that the central bank will fully offset the shock, then market expectations of inflation may not change. If the central bank looks at the inflation expectations and concludes that there is no threat to aggregate demand, the central bank may end up not offsetting the shock, and the markets fall in response.

The most recent example of this in U.S. financial markets is the Fed taper. Before the Fed taper talks, the general expectation was that quantitative easing would continue into the indefinite future and that there would be no premature tightening. As a result, the stock market seemed very resilient because there were expectations of strong growth conditional on Fed easing. The Fed misinterpreted these expectations as independent of the Fed's policy of QE and decided to tighten.

However, fiscal authorities can also face the same circularity problem. If an economy is highly dependent on government spending, then real economic conditions may be determined conditional on expected future fiscal easing. And if the fiscal authority sees the strong current economic conditions as a justification for austerity, then this too may cause a fall in growth in the same way that a premature monetary contraction can slow growth.

The Chinese government is currently facing this fiscal policy circularity problem. In an interview on June 18th with the IMF mission chief for China Markus Rodlauer, he notes that high frequency data such as retail sales, investment growth all point to moderate growth. Even thought the PMI may have faltered a little bit, it's well within historical ranges. Rodlauer takes this and makes the conclusion that there's really no need for stimulus. 

While he may be right, it is also likely that the Chinese government could fall into a fiscal circularity problem. Especially since Chinese fiscal policy has the ability to reallocate a large amount of resources, much of business is conducted on the basis of expectations of future government policy. Under these conditions, concluding that economic conditions are strong on the basis of high frequency data may cause the fiscal authority to be too sluggish in responding to a slowdown in growth.

Casting and Melting with Paired Data

Today's post is not about economics, rather it's a note from an R programming struggle that may be helpful for fellow undergraduate researchers.

I'm often testing forecasting models, and what this ends up creating is a bunch of "forecasted" variables that are paired with the "actual" values. R has fabulous faceting capabilities, and I have often wanted to reshape the data in a way where the category of forecasted variable as an identifier, and then two columns that list the forecasted and actual variables. In other words, if the code starts from something like

       aAct      aPred       bAct      bPred id
1 1.2076384 -0.6735547  1.4994464 -1.0691975  1
2 0.4999706 -0.7188215 -0.3601551  0.7224729  2
3 1.0340859 -0.1108304 -0.5941295  0.5027085  3

And I want to convert it where one column has an id, another one identifies whether I'm forecasting a or b, and a third column that has the forecasted value, and then a fourth column with the actual value.

The procedure in R involves "melting" the data frame and then "casting" it. Melting is rather simple -- you provide a set of identifiers, and then the data frame is melted down to only that identifier, the values, and another indicator variable that tells you what the value is supposed to represent. In the above example, if we let df be the data frame described above, I would run:

df.m = melt(df, id.vars = 'id')

   id variable      value
1   1     aAct  1.2076384
2   2     aAct  0.4999706
3   3     aAct  1.0340859
4   1    aPred -0.6735547
5   2    aPred -0.7188215
6   3    aPred -0.1108304
7   1     bAct  1.4994464
8   2     bAct -0.3601551
9   3     bAct -0.5941295
10  1    bPred -1.0691975
11  2    bPred  0.7224729
12  3    bPred  0.5027085

Now I need to "unmelt" part of the data frame to get the forecast/actual pairings. In R, this is known as casting and I know that I personally had a pretty hard time decoding the documentation. The function goes along as

cast(df.m, <IDENTIFIERS> ~ <VALUES>)

The second part is known as the casting formula and is the part that I have struggled with. But in its most simplest form, the casted frame will look like something with all the identifiers added together as uniquely identifying units ,and then the <VALUES> variables being the labels for the actual value column. If that sounded confusing, I apologize. Perhaps solving the example would help.

First, I need to find a way to identify whether a row is looking at a or b, and whether it is a forecast or an actual variable. So I first create these variables:

df.m$var = substring(df.m$variable, 1, 1)

df.m$type = substring(df.m$variable, 2) 

Which gives me the data frame:

> df.m
   id variable      value type var
1   1     aAct  1.2076384  Act   a
2   2     aAct  0.4999706  Act   a
3   3     aAct  1.0340859  Act   a
4   1    aPred -0.6735547 Pred   a
5   2    aPred -0.7188215 Pred   a
6   3    aPred -0.1108304 Pred   a
7   1     bAct  1.4994464  Act   b
8   2     bAct -0.3601551  Act   b
9   3     bAct -0.5941295  Act   b
10  1    bPred -1.0691975 Pred   b
11  2    bPred  0.7224729 Pred   b
12  3    bPred  0.5027085 Pred   b

Now I can cast the frame. In this case, I would use the formula

df.mc = cast(df.m, id + var ~ type)

This is how you interpret the formula. Id + var means that every observation is uniquely identified by it's id code and the variable we're forecasting -- a or b. Then "type" on the right side represents the new variable names that will be filled by the values.

Hope this is useful to others so they don't end up spending hours agonizing over the issue as did I.

More on Growth and Convergence Within Countries

In my last post on China, I touched on the issue of Chinese growth by showing a graph with the distribution of Chinese per capita incomes by province, and arguing that there is a strong convergence story pushing China towards more growth. In the comments, Tamar makes a note that many countries do not converge. For example, per capita incomes in Mississippi and Connecticut differ by a factor of about 2, even though the United States is a relatively developed Country. He also suggested that I take a look at Brazil. And so I did. I took a look at the distribution of  province per capita income divided by country per capita income for three emerging market economies: Brazil, Mexico, and China, and found that indeed, they were quite close!

I wondered if it was because I didn't weight for populations, so I downloaded some Mexican population data from their government's website. I didn't have time to do Brazil, but even comparing China and Mexico I found that the distributions were quite similar.

On first pass, this bodes poorly for a convergence hypothesis.

But let's think back to the Solow model. We should only observe convergence in income levels if technologies and savings rates are all identical. But it's entirely plausible that these can differ across provinces, and that they differ for extended periods of time. Therefore a better metric to evaluate convergence is not whether they converge in levels, but rather if they converge in growth rates. In the Solow model, at the steady state, all countries grow at a rate equal to the rate of population growth plus the rate of technological change. If they're all bound together (eg if they're all large counties in one country), then g should be similar across them, and demographic trends typically do not differ hugely among provinces in the long run.

So if we look at growth rates, now we see convergence at work. As a technical note, I only had data for Mexico from 2003 to 2010. So I got the ratio by exponentiating the 7 year ratio by 10/7. 

So even though Mexico and China have similar distributions in terms of their with country income levels, they have widely different distributions for growth. Therefore I stand by my original belief that China still has a lot of long run growth potential to go as the poor provinces catch up to the rich.

China's Provinces and why National Data can Mislead

Close your eyes and think of China. What do you see?

If you were like me, you saw a large metropolis filled with high rise apartment buildings, inked with chronic air pollution, humming along to the sounds of millions of residents getting through their days.

I believe this is also the image many economic commentators have in their minds when they talk about an upcoming "Chinese" slowdown. But what I want to do in this short little post is to demonstrate why thinking this way neglects one of China's most important quality: its size.

China has a total of 1.34 billion people spread over 23 provinces, 4 municipalities, and 5 autonomous regions. Individual provinces in China can have as many people as entire countries. The coastal province of Guangdong has a population of 105 million -- just shy of Mexico's 112 million and far exceeding every country in the European Union. Sichuan, an inland province (known for its spicy food), has a total of 80 million inhabitants -- larger than the entire Western Untied States combined. In this sense, it's better to think of China as a collection of smaller countries united under a currency union called China, and not as a uniform economic entity.

For example, consider the following map from Wikipedia showing per capita income by province.

As can be seen, there are vast disparities in income. Whereas the coastal provinces are quite rich, the inland ones are quite poor. However, the chart understates these differences because it uses a log color scale. Below is a histogram of the 2012 per capita income and population statistics pulled from the China Data Center associated with the University of Michigan.

GDP per capita in Shanghai was 85,000元 whereas GDP per capita in neighboring Anhui was only  28,792元. Translated into market exchange rates this means an average GDP per capita of $13,848 in Shanghai and only $4690 in Anhui. If we take the Solow model seriously, what this suggests is that there is a massive potential for convergence within China. Even if the inland provinces do not face as favorable conditions as the coastal provinces did when they got rich, do you really expect the 80 million residents of inland Sichuan to stay at 60% of coastal Guangdong's income forever? Especially since China does do so much manufacturing, Dani Rodrik's work on unconditional manufacturing convergence suggests that these poorer provinces will inevitably partially catch up with the richer provinces. There's just not enough income for them to get caught in a middle income trap.

There is also no systematic relationship between population and income. No matter the combination of big or small, rich or poor, there is a Chinese province that fits the description.

Recognizing this heterogeneity also provides a good reason for why looking at China's GDP per capita statistics provide an overly rosy picture of China's wealth and an overly dour prospects of China's future growth. Because there are a few provinces that are now somewhat rich while most provinces are still very poor, mean GDP per capita for the nation does not accurately represent the plight of most provinces. You can see this by the fact that most provinces in the above scatter plot are below the regression line that approximates the mean level of GDP per capita. As a result, we underestimate the role convergence has to play in bringing more Chinese economies out of poverty and therefore underestimate the true growth potential that China has.

Bottom line is that "turning point" arguments that fail to consider the subtleties of individual provinces will lead us astray. Too often, we associate China with middle income images of massive apartment complexes, where in reality much of China is still very poor. Any serious evaluation of where China is going requires careful consideration of how we think growth in individual provinces will evolve. And based on the provincial data, I am quite optimistic.

A Market Monetarist Approach to the Interest Rate Puzzle

What’s going on with real rates, inflation breakevens, and the stock market? From the beginning of 2010 to the end of 2012, these three variables have affected each other in a predictable way. Higher inflation breakevens pushed up the stock market as they served as a sign that aggregate demand was rising. Growth in real rates was associated with increases in the stock market as the real rates served as a predictor of future growth. However, these relationships have broken down in this first half of 2013. In this post, I aim to explain why. By combining movements in market data with traditional economic theory, there is convincing evidence that the recent change is due to a positive aggregate supply shock, and therefore bodes well for economic growth looking forward.

This post will proceed in three acts. In Act One, I introduce some work that has already been done on this question. In Act Two, I present a new approach to process the market data and the theory that justifies the observations. And in Act Three, I address any residual concerns. Let us now begin.

Act One -- The Work that Has Been Done

Recent trends in financial markets since 2010 are summarized below. In it we have the movement in the 10 year real interest rate, the 10 year inflation breakeven, and the SP500. During the 2010-2012 time period, the 10 year inflation breakeven was very tightly correlated with the SP500, and if you squint you will notice that increases in the 10 year treasury yield also were correlated with increases in the SP500. However, this seemed to reverse itself starting in 2013. Even as inflation expectations were falling, the SP500 still gained steady ground. Also, when the 10 year real interest rate spiked in recent weeks, we saw a temporary fall in the SP500.



Evan Soltas has documented the breakdown of the interest rate relationship. There are two signals communicated by a rising rate. First, it could be a signal of stronger future growth -- which should send the SP500 up. On the other hand, it could be a sign that monetary policy will be too tight -- which should send the SP500 down. By looking at 90 day rolling correlations between the daily percent change in the 10 year treasury yield and the SP500 stock index, we can tell the difference. Evan has observed that the correlation coefficient between the two changes is quickly approaching zero. According to him, this signals that “over the past 90 days, monetary tightening has been as important to rates as has been macroeconomic strengthening”. The June survey of primary dealers further confirms this hypothesis.

Brad Delong and Matt Yglesias have both come into this debate on Evan’s side, arguing that the Fed has been engaging in a stealth monetary policy tightening. To them, these trends are signs that growth could suffer again in the upcoming months as the Fed decides to tighten too early.

On the other hand, I have looked at the relationship between inflation breakevens and the SP500 and believe what we’re really looking at is a positive supply shock. I find that even though 2013 has been characterized by falling inflation breakevens alongside a rising SP500, marginal increases in the TIPS spread still have a positive effect on equity prices. The only difference is that the SP500 seems to have a higher trend growth level -- an alpha with respect to inflation, if you will. I interpret this as an expectation of higher output at every level of inflation. I identify this with a textbook increase in aggregate supply, and thus argue against the monetary tightening hypothesis.

Act II - Another Look at the Data

One unfortunate oversight of the analysis Evan and I have each done is that we don’t fit our stories together. He says tightening, I say aggregate supply, and we each point to our individual data. But an open question remains: how do our theories explain the other person’s data?

To try and estimate this, I roll with Evan’s calculations, but with slight modification. Instead of calculating correlation coefficients, I instead compute rolling regression coefficients. I look at week to week changes in inflation breakevens, the 10 year TIPS yield, and the SP500. For each week I compute regressions of percent changes in the SP500 against percentage point changes in the TIPS yield and inflation breakevens for the past 26 week window. The regression slopes measure the response of the SP500 to either interest rate changes or expected inflation. It corresponds loosely to the correlation coefficient Evan calculates. The regression intercept measures the “intrinsic” trend growth of the SP500, independent of interest rates or inflation. By looking at these coefficients in context, I will try and construct a more holistic vision of what the financial markets are trying to say.



First, let us take a look at the right side panels which describe the responsiveness of the SP500 to expected inflation. In some sense, changes in the SP500 represent changes in expected future nominal GDP. Therefore, when we look at the relationship between inflation expectations and the SP500, this serves as a proxy for the relationship between inflation and nominal GDP.

In my view, the spike in the TIPS breakeven intercept is a smoking gun for a positive aggregate supply shock. Think about what the higher intercept means. The regression is of changes in the SP500 against changes in the TIPS breakeven. Therefore an increase in the intercept means that the SP500 grows faster for every level of expected inflation. This effect is quantitatively important as well. In comparison to the 6 months ending 2012, the intercept for the past 6 months suggests that the SP500 has kicked it up from about 0% weekly trend growth that is independent of inflation expectations to about 0.8%. Meanwhile, the slope for the breakeven-SP500 relationship is still positive. This all suggests a more permanent aggregate supply shock is driving the intercept up, whereas day to day aggregate demand shocks keep the slope positive. A diagram of this is shown below.

However, careful readers will note that you can get “more output at every price” from a story with a structural shift in aggregate demand with marginal shocks coming from aggregate supply. However, this hypothesis fails on two counts. First, if marginal changes in inflation reflected changes in aggregate supply, not demand, then because aggregate supply shocks send prices in the opposite direction of output, we should expect the TIPS breakeven slope to be negative. Second, the AD story does not match up with the changes in levels. As I showed above, inflation expectations have fallen while the SP500 has risen. If there were a large aggregate demand shock, then we should have seen both the SP500 and TIPS breakeven rise in levels. Therefore, a positive aggregate supply shock provides the most natural interpretation for the right hand panels.

Now comes the out of sample test. Can an aggregate supply shock explain the low slope and moderately higher intercept in the SP500-real rate relation? Absolutely.

To see how, I appeal to a version of the IS-MP (Investment Savings, Monetary Policy) model, pictured below. In the diagram, nominal GDP growth is on the x-axis and the real interest rate is on the y-axis. The IS curve is the standard IS curve from intro macro. It describes various combinations of interest rates and nominal GDP levels that give equilibrium in the goods market. At lower levels of the real interest rate, people want to hold onto less money and consume more goods. This results in higher levels of nominal GDP and a downward sloping curve. The MP curve is slightly different because it describes not equilibria but a central bank reaction function. At higher levels of nominal GDP, the Fed sets higher a higher interest rate in order to prevent rapid inflation. These two curves now give a unique equilibrium characterized by an interest rate and a level of nominal GDP.

Now what happens if there is a supply shock? The increased productive capacity, on first approximation, has no effect on the IS curve. To see why, suppose the monetary authority does not react. Then because a supply shock leaves nominal GDP relatively unchanged, then the IS curve should not move. However, because the Federal Reserve is an inflation targeting central bank, the MP curve shifts down. Now that every unit of nominal GDP consists of more real growth and less inflation, monetary policy becomes easier. Therefore, the new monetary policy curve will look something like MP(2) in the picture above.

This matches two more details from the regression.

First, the downward shift of the MP curve means that at every interest rate you observe more output. This matches the somewhat higher regression intercept on the interest rate graph.



Second, since the MP curve is moving, we should expect a weaker correlation between interest rates and output. This is illustrated above. If the MP curve is held constant while the IS curve shifts back and forth, then we will observe a strong correlation between interest rates and output, as shown by the blue line. On the other hand, if the MP curve is moving to MP(2) at the same time, we may end up observing the red dots and finding that the correlation drops. We also should expect this correlation confusion to be a bigger deal for the monetary policy shift than for the aggregate supply shift. As Bernanke is finding out, shifts in MP are linked to relatively unstable market expectations whereas a positive AS shock from something like oil discoveries is much more predictable. The theory behind this explanation of the fall in the correlation is illustrated in the sketch below, and it is actually exactly what we observe in the markets during the first half of this year.

The final step is to get the higher interest rate from the taper, and this can be seen as just the effect of a slight Fed tightening along with a slight rightward shift of the IS curve as business confidence requires. In the end, you have higher growth, higher rates, even though the Fed has tightened (as per the Dealer survey) relative to where it was before.

Act 3 -- Addressing Additional Concerns

While I believe the above story is the one most consistent with the regression data, there are always additional concerns.

Most importantly: what is the positive supply shock? I believe the most plausible supply shock could be the further discovery and development of unconventional oil and gas reserves. Therefore when compared to the counterfactual of perpetually rising oil prices, the new discoveries makes it easier for policy makers to respond to energy shocks and improve the economy’s productive capacity.

An important note is that a rise in oil prices, when it occurs alongside a rising SP500, does not contradict the aggregate supply hypothesis. An aggregate supply shock is characterized by a general fall in inflation as output rises. But if aggregate demand is moving at the same time, we could end up observing higher prices with even higher output. Therefore we identify an aggregate supply shock by seeing higher output *for any given level of inflation*. And this is precisely what we see from the rolling regressions.

Also, I have somewhat of a harder time explaining the past movements in the intercepts and slopes. Fortunately, the intercepts seem to move up and down together, whereas the slopes do the same. Moreover, the intercepts often go in opposite directions when compared to the slopes. This suggests that supply shocks may be more recurrent than we are led to believe.

Others may criticize the above approach as too ad-hoc. While to some extent, it certainly is, I believe I have done justice to the spirit of the AS/AD and IS/MP models. Furthermore, if you break down all the layers of abstraction and ad-hoc econometrics, the story is quite simple:

The massive increase in U.S. petroleum resources has expanded aggregate supply, allowing the economy to attain higher levels of output at every level of inflation. This serves as a massive tailwind for equity markets that no longer depend on aggregate demand inflation to grow. This requires a muddled monetary policy adjustment -- reducing the previously observed correlation between interest rates and growth. Nonetheless, the aggregate supply shock has increased trend growth, making the fluctuations in interest rates matter less.

Fin.

The Reach for Real Bills

Awash with liquidity and starved of paper, must financial markets slip out of control? This is the central question behind the “financial stability” argument against additional monetary easing. According to this objection, the zero bound on interest rates means that the Fed’s easing can do little for the real economy, and the cash created by open market operations just fuel a speculative excess termed a “reach for yield”. I have addressed one reason why this theory is incorrect. If QE indeed spurred a reach for yield, then the taper talk should have reversed this and caused a flight to safety. Yet after the taper dust settled, we saw cyclicals rally strongly with safe assets falling -- indicating that QE was likely encouraging healthy risk taking and not an anomalous reach. However, this evidence primarily came from equities. In this post, I want to take a different approach to expand the scope of my argument against financial stability concerns. I will start with some monetary history and discuss why thinking in terms of financial stability can be very misleading. In short, adopting financial stability approach to monetary policy is unwise and will likely worsen both the business and financial cycle.

First, let’s consider the motivating evidence for the financial stability position. Below is a chart prepared by UM alumni Naufal Sanaullah charting the loan deposit gap into US commercial banks. According to Naufal, this shows that the usual lending mechanism that we learn in intro macro doesn't work any more. No more loans are going out, and therefore nothing makes it to the real economy. And while the real economy is unaffected, this domestic savings glut drives a reach for yield as banks still need to pay their depositors.



If this theory is correct and monetary policy is completely ineffective, the Fed should taper earlier. If the costs to financial markets are great enough, and if the benefits to real economies are small enough, it may be worth it for the Fed to fumigate any excess risk in markets by raising interest rates.

Thinking in terms of financial stability may seem novel, but the Federal Reserve actually had the same debate during the Great Depression. Julio Rotemberg, in his recent paper for the NBER monetary policy conference, does a wonderful job summarizing the literature on the thought process of the Fed at that time.

Friedman and Schwartz (1963) stressed instead the substantial declines in the money supply that followed. These were, in part, the result of the Fed’s refusal to lend to banks subject to runs. In addition, and in spite of the exhortations of various Federal Reserve officials at various times, *the Fed resisted embarking in large-scale open-market purchases to offset the declines in banking.8 Under pressure of Congress, such a program was started in April 1932, though it quickly ended in August of the same year. This was rationalized on the ground that conditions were “easy” since there were ample excess reserves. Some officials thought the increase in excess reserves (and reduction in borrowing from the Fed) proved that the program was ineffective.9* 

Given subsequent developments, it seems likely that some members also viewed excess reserves with fear. As excess reserves accumulated in the mid-1930s these fears were openly discussed, and Friedman and Schwartz (1963, p. 523) quote extensively from a 1935 memo that clarifies their nature.* In effect, the Fed worried that banks would use these funds for speculative purposes that would ultimately be costly. *Or, as the 1937 Annual Report put it, the Board feared “an uncontrollable increase in credit in the future.”10 *These concerns were sufficiently intense that the Fed raised reserve requirements by 50% in August 1936. Further increases in 1937 left them at double their 1935 values (Meltzer 2003, p. 509).*

If you look closely, the parallels to the Fed’s dramatic QE policies and current financial stability concerns are uncanny. In both stories, the recession was identified as the result of speculative excess. In response to the crash, both times the Federal Reserve embarked on a program of monetary easing. However, in both instances excess reserves failed to budge, and this was interpreted as a sign that banks just didn’t want to lend -- the Fed was pushing on a string. Finally, as excess reserves persisted, the threat of “speculative purposes” was used to bully the Fed into tightening. The key difference between now and then is that we have a Fed that recognizes its role in supporting the real recovery. Those in 1936 were not as lucky.

Why did the Fed go on such a destructive path in the 1930’s? Rotemberg identifies the tightness of policy as a consequence of something called the “real bills doctrine”. Under the real bills doctrine, the Fed saw its role as providing credit so that there was enough, and no more, credit to invest in “productive uses”. Since the Great Depression was preceded by a speculative stock bubble, then Fed officials put a premium on making sure credit was put to “productive uses”; The real bills doctrine was the result. According to this doctrine, monetary policy should tighten in recessions when demand for credit falls so as to make sure what credit remains is put towards productive uses. Conversely, monetary policy should ease in booms because firms are looking to find credit to fund their projects. In other words, the real bills doctrine prescribed a procyclical monetary policy.

This goes to show that we need to avoid framing effects when thinking about monetary policy. Because the Great Depression was the result of an equity bubble, then the economists of the day were so concerned about bubbles that they pursued destructive monetary policy. It is just as important to not make the same mistake today. As the real bills doctrine shows, using the tools of financial economics to solve monetary problems can be very destructive.

In particular, the concern about excess reserves or a loan-deposit imbalance comes about from ignoring general equilibrium. Walras' law states that the value of excess demands add up to zero across all markets in an economy. So if there is a lack of demand in goods, it must be the result of an excess demand for money that goes into savings. But if the interest rate is low enough, it may no longer be worth it to hold onto the money as savings and people will spend it. In the limit, if people knew that all of their cash would disappear when the next day started, they would certainly spend today. There must be a real interest rate, perhaps negative, that would make people want to give up enough money to equilibrate the goods market. This conclusion now recasts the question to whether that negative rate is attainable. Once you can reach any arbitrary rate, then the money markets and good markets are sure to equilibrate.

Of course if the Fed was stuck at the current interest rate it could never attain the negative rate. But that’s where forward guidance comes into play. What forward guidance allows the Fed to do is pin down the future price level -- even if there appear to be no tools right now. This is the well known escape clause in Krugman’s original analysis of the liquidity trap. If the Fed can commit to a future policy path, the zero lower bound no longer matters.

To get a more intuitive feel for this argument, you should think in terms of an observable Fed policy rate (r) and an unobservable Wicksellian, or full employment, rate (w). The full employment rate is so named because it is the interest rate at which all resources are fully employed. In this example, I set both interest rates to be nominal, so r cannot be lower the zero. At any given instance in time, the stance of monetary policy is determined by where the policy rate, r, is relative to the Wicksellian rate, w. If the Fed rate is higher than the Wicksellian rate, the Fed is tightening. If it is lower, the Fed is easing. Dynamically, the Fed's policy stance is determined by the blue area minus the red over all time.

This gives a natural interpretation for why forward guidance works at the zero lower bound. Even though the Fed’s rate, r, is stuck at zero and is currently above the Wicksellian rate w, the Fed can still generate inflation by promising to keep Fed policy easy in the future, even when the Wicksellian rate rises. This then can move the economy to a different equilibrium. With higher expected inflation, the nominal Wicksellian rate rises since means people are willing to part with their money (read: have no excess demand for money) at higher interest rates. As a result, even though the Fed is constrained right now, it still has power over the future policy path. This goes to show that the zero lower bound is not a serious reason to discount the Fed's ability to conduct monetary policy.



To get back on track, the Fed must commit to keeping rates low until the price (or nominal GDP) level is back to trend. On the other hand, if the Fed were to raise interest rates now, this would collapse expected inflation, lowering the Wicksellian curve and knocking the economy into a low output, low interest rates environment. So even if you think the low rates environment is causing financial distortions, the only way to get higher rates in the future and to solve the apparent financial distortions of low interest rates is, ironically, to promise to keeping short rates low now.

The financial stability view gets off track because it ignores general equilibrium effects. In partial equilibrium analysis, when there's an excess stock of something, such as bank reserves, the natural response is to cut supply. But this is misleading analogy for bank reserves, because an excess supply of bank reserves actually represents an excess demand for money. Therefore the proper response is to maintain lower rates and not prematurely tighten.

Therefore the real bills/financial stability doctrine fails for three reasons. First, it identifies excess reserves as the result of reduced borrowing that the Fed cannot control, whereas the excess reserves actually are symptoms of an excess demand for money that easier monetary policy can address. Second, this misdiagnosis means we are left thinking the Fed is powerless, whereas the Fed can pin down the price level through forward guidance. Third, it ignores the general equilibrium relationship between money and goods. By prematurely raising rates, this actually depresses interest rates in the long run and worsens the excess demand for money. Bottom line? Worrying too much about financial stability concerns can exacerbate the business cycle and actually prolong a period of low rates. Instead, the Fed should keep its eyes on the real economic prize, and keep financial decisions separate from its monetary ones.