IMF Book Forum: Beating the Business Cycle: Can Turning Points in the Economy be Predicted?
People's Republic of China and the IMF
India and the IMF
Free Email Notification
Transcript of an IMF Book Forum|
Beating the Business Cycle: Can Turning Points in the Economy be Predicted?
Thursday, June 23, 2005
International Monetary Fund
MR. LALL: Good afternoon and welcome to this IMF Book Forum. My name is Subir Lall, and I will be moderating this discussion. Today's book, by Lakshman Achuthan and Anirvan Banerji of the Economic Cycle Research Institute (ECRI), is "Beating the Business Cycle". I personally found it very interesting to read, and I also went to their website and toured around there a little bit, as much as I could without actually paying for anything.
But, you know, some of you might feel the urge to bring out your credit cards and move further in than I did.
The topic of the book is close to the heart of most IMF economists, but also to people in general in the economy and to politicians of course. It's important to understand why business cycles occur, how to recognize them ahead of time, and what to do about them. It isn't easy to tell when a recession begins and when it ends and how to predict it. It's a bit like the flu, which is often confused with a cold. You're not sure whether you have it or not, and you might be taking the wrong medicine. And it affects pretty much everybody at some point or the other. So in market economies, business cycles are a key underlying fact of life, even though there have been many periods when their death has prematurely been announced. But as this book makes clear, business cycles remain a fact of life and will so for at least the foreseeable future.
Most forecasters of business cycles have a very dismal record in forecasting downturns and upturns in the economy. Economic Cycle Research Institute is one of the few which has actually had a very stellar record. They have been making pretty bold calls and going against the conventional wisdom. So far their record has been one of the most impressive, and has been written up in the press as well as talked about in policy circles.
Now, why should we care about business cycles? This book does an excellent job in actually trying to put forward the idea that we should care and that there is an easy way-or at least not a very complex way-to try to understand them. So using an analogy from the book, if you're driving a car, you don't need to know how the engine is constructed, but you certainly should be able to read the gauges and take your signals from them to improve your own ability to drive successfully.
It's rare that a book about a technically complex subject does not fall into the trap of writing in very arcane terms, which most people will lose interest in rapidly outside of a core group of researchers. At the other extreme, there are often books which take a complex subject and dumb it down to make it easy to understand but sometimes don't do justice to the topic, though they may sell well and they have catchy covers. This is one of those rare books which I find personally has bridged the gap in taking a very complex subject and making it interesting to the readership, a wider readership than purely people who are in the profession.
Another interesting facet of this book is that ECRI has a very internationally-focused outlook. They have indices for more than 20 advanced and emerging market economies, and they recognize the interactions among them. So it's not merely U.S. centric, though they have about 20 indexes for the U.S. alone. But their work goes well beyond the U.S. and recognizes interactions and developments that may happen elsewhere in the world. They use this to analyze, for instance, the boom period in the 1990s and how global factors helped explain the cause of that boom.
I'm pleased to have with us here Anirvan Banerji. He helped Geoffrey Moore, the noted pioneer of business cycle forecasting, to found the Economic Cycle Research Institute after working with Moore for more than ten years at Columbia University and helping develop many of the well-known indices of ECRI. Some of the leading indices that he has helped develop are the weekly leading index (WLI) and the future inflation gauge (FIG). He has published extensively in academic journals and also advised central banks and international organizations such as the Asian Development Bank. He is a member of the OECD Expert Group on Leading Indicators, and Forecast Chair of the Forecasters Club of New York. He's the Director of Research at ECRI and has worked with Lakshman Achuthan, a colleague at ECRI, in producing this book.
Anirvan will speak for the first 30 or 35 minutes, followed by our two discussants, after which we welcome audience participation. Anirvan, it's all yours.
MR. BANERJI: Thank you very much for that very kind introduction, and thank you all for coming. I'm actually a bit surprised to see this large an audience. In fact, Fred Joutz was just telling me that we both he and I made an incorrect forecast about the size of the audience here.
So thank you.
Last week, I was at a forecasting conference where a well-known econometrician was comparing the accuracy of various approaches ranging from Bayesian to Markov switching models, threshold models, VAR models, and all that. And while some of these models perform better than others, they were all clearly outperformed by the Survey of Professional Forecasters. The superiority of such consensus forecasts is well known. And yet, according to a study of consensus forecasts by the IMF's very own Prakash Loungani, their "record of failure to predict recessions is virtually unblemished".
And so it was when the last recession arrived. The Economist described the consensus: "In a survey in March 2001, 95 percent of American economists said there would not be a recession." Of course, a recession had already begun. ECRI is perhaps the only organization, to quote The Economist again, to give "advance warning of each of the past three recessions; just as impressive, it has never issued a false alarm."
Well, you know, in baseball a batter is considered pretty good if he manages to hit above .300 or 30 percent over the season. After Barry Bonds, one of the greatest ever, was featured on the cover of Sports Illustrated magazine in 1993 his batting average plummeted 40 points. There is something of a Sports Illustrated jinx, a demonstrable misfortune or decline in performance following a cover appearance roughly 37 percent of the time.
Now, there are two points I want to make with this analogy. First, forecasters would be lucky to bat .300 or even .250 when predicting recessions. The second is that we at ECRI are in a possibly terrible situation with The Economist pointing out that we are the only group that has been right 100 percent of the time, batting a thousand in predicting recessions. We just have to pray that there isn't a jinx associated with The Economist that will now spoil our record.
Now, consensus forecasts are pretty good most of the time when the economy is trending in one direction or the other. It is when the trend changes, when you approach a turning point, that the forecast error balloons. Why are turning points so hard to predict? The reason is that most forecasting models tend to have an element of persistence or an extrapolative quality to them. So the turning point error is the big error. And consensus forecasts, which are known to be among the most accurate, systematically make large errors in the vicinity of cycle turning points.
So to improve the forecasts around turning points, we have designed leading indicators—indexes that are designed to specifically lead at the turning points of the cycle. There are some subtleties. For example, there are asymmetries. The leads at peaks are generally longer than leads at troughs. There are mismatch amplitudes and so on. But without getting into all those details, the simple point is that we use a leading index to predict turns. But if we are simply using a leading index approach, which is nothing new, frankly, how can our results be so much better?
To, answer this, I need to explain some of the details of the evolution of the work of Geoffrey Moore and ECRI, some of which you may know and some of which may be a bit new. At the most basic level, one point of departure between our approach and those of many econometricians is our belief that recessions are triggered by a combination of endogenous forces and exogenous shocks, not unpredictable exogenous shocks alone. Otherwise, there would be no question of successfully predicting turning points. But we have shown time and again that it's possible. In fact, the leading indicators are a measure of these endogenous forces.
Early business cycle research was pioneered by people like Wesley Mitchell in the beginning of the 20th century. One landmark in 1927 was arriving at the classical definition of business cycles. Then working with Arthur Burns, Mitchell came up with the very first leading indicators of cyclical revival, as he called them. Burns and Mitchell started things off, and that's about when Moore joined them at the National Bureau of Economic Research. Geoffrey Moore made some important advances on their work, and prominent among them is that he created the first list of leading indicators of revival and recession. That was back in 1950. He also developed the composite index of leading indicators. A lot of people think it's Burns and Mitchell. It's not. It's really Moore and Shiskin in the mid-1960s.
Most of you probably knew much of this, but there are some things here that you may not know. For instance, since then, in the last 40-plus years, Moore and his colleagues like Philip Klein, who is still working with us, pioneered the application of his approach to international economies in the early 1970s and 1980s, and created separate leading indexes for inflation and employment. Now, ECRI's forecasting approach builds on all of this work of Geoffrey Moore's, not just Burns and Mitchell. And I have to say that what we've done in recent years is build on that: standing on the shoulders of giants, that's really the source of our success.
Now let me outline the features that set off ECRI's work from others that I believe is responsible for our distinctive forecasting record. What are the points of departure? I will focus on a few. First is the question of optimization versus robustness. The second is the multidimensional framework we use. The third is the breadth of international coverage. And fourth, there's a question of independence, not statistical independence but the independence of our institution. Because of all these factors, what our approach has allowed us to do is to diverge from the consensus as we approach turning points.
1. Optimization vs. Robustness
Let me start with an anecdote. Back in the late-1980s, a pair of well-known econometricians created a sophisticated recession probability index. But the index really failed to predict the 1990 recession. I remember some years later at a seminar somebody asked one of them: "So how come it failed? Why didn't it work in predicting recessions even though you followed a fairly sophisticated approach? And the answer he gave was very revealing. He said: "Parameter drift."
Now, if you think about that, they had estimated these parameters over a given time period, and the values had changed since then. If that's going to happen, how on earth are you going to predict anything, especially something like turning points? That is an inherent danger of an approach of trying to optimize the fit of the model. The standard approach is very much trying to optimize the model for a given time period, for a given economy, et cetera. Our approach is not so much about optimizing what would have worked best over a given period. It is more a question of robustness. Will these indicators, will the system hold up? That's the question.
To illustrate this issue, let me go back to the original list of leading indicators created by Moore back in 1950. This was the first list of leading indicators of revival and recession. How did Geoffrey Moore arrive at that list? Here's the key: He started with an understanding of the basic drivers of the business cycle, building on the work of Mitchell and Burns and others. It's important to understand that there was a clear conceptual underpinning to this work, and based on that understanding, Moore came up with a list of all the indicators where you should expect to see the early signs of a turn in the cycle.
And then he subjected them to an empirical test based on the U.S. data he had at hand at the time, which was U.S. data from 1870 to 1938. There were eight leading indicators of revival and recession that made the cut. If you look at these average leads, you find that, on average, they would have four-month leads at peaks and close to that at troughs. That's what Moore was looking at back then.
A little over a decade ago, Moore asked: what have these indicators done for us lately? And in answering that question, he rebutted a well-known critique that had been lobbed at his mentors, Burns and Mitchell, which is that their work was essentially measurement without theory. Moore always knew that their work was based on as much theory as the evidence would support without making premature leaps of faith to build precise but falsifiable mathematical models. The critics of Burns and Mitchell's approach made a critical mistake that Sherlock Holmes for once warned against. If you remember The Scandal in Bohemia, he said, "It's a capital mistake to theorize before one has data. Insensibly, one begins to twist facts to suit theories instead of theories to suit facts."
At that point in the development of business cycle theory and business cycle research, we simply didn't know enough to formulate a serious model of the business cycle. The fact is we still don't, okay? But what Mitchell said back in the '20s--and that's the research project that we're still carrying on at ECRI-is: "Let us look at this phenomenon in every setting where it has appeared. Let us try to understand what has gone on. Let us observe first what has happened. Then start the task of theory building." We still are embarked on that long project. And so Moore titled his paper, "An Answer to Measurement Without Theory."
Now if you look at the performance of the same indicators in the second half of the 20th century in the U.S., what you find is that they are still holding up. A little longer lead on average at business cycle peaks, a little shorter lead at troughs than before. But they are still leading. So, remarkably, the leading indicators which correctly predicted turning points in the post-Civil War Period kept doing that in the second half of the 20th century. And I don't have to elaborate on the profound structural changes the U.S. economy saw in that period.
These indicators were robust enough to keep working not only in the U.S. but also in ten other economies, including the G-7. The leads held up - there was about a four-month, five-month lead in these other countries on average. Now, stop and think about this for a minute. We are making a pretty bold claim here. What we are saying is that the same indicators that anticipated turning points in the post-Civil War U.S. economy continued to work in late 20th century Germany, South Korea, New Zealand, and, of course, the U.S. I'm unaware of any other forecasting approach to show this degree of robustness over time and across countries.
To repeat: the reason these indicators work over space and time, so to say, is that they capture the key drivers of business cycles.
2. Multidimensional framework: The State of the Art
At the same time, I also want you to appreciate that what I have talked about thus far is hardly state of the art. The framework we use today to monitor the economy goes way beyond this. This framework is a major evolution of the work that Mitchell, Burns, and Moore began. We call it the `monitorist' framework.
What I mean is by saying we are `monitorists' is that we monitor the economy. We don't so much forecast it, but we monitor where we are in the business cycle and how the indicators are evolving.
Let me explain. We today use a multidimensional framework. Sure, component selection, which is what we have been talking about so far, is essential. But so is grouping. Here we start with a view that there are three key aspects of the economy. While they are related, they have different enough timing of their respective cyclical turns that we believe they warrant unique attention. As a convenient schematic, we use a cube which has inflation, employment, and growth as the three facets. We don't literally mean to imply that these dimensions are independent.
Now, what's interesting, by the way, is that quite independent of this, the econometricians Stock and Watson arrived at what I think are kind of similar conclusions. In their paper on what they call diffusion indexes, they essentially did a factor analysis on a large number of economic variables. You find that the first factor seems to be related essentially to production or growth; the second related to prices, interest rates, etc; and the third to employment. Of course, the labels in any factor analysis are always sort of fuzzy. You never know exactly what to call them, but that is certainly what you'd see if you looked at it carefully.
Now, let's sort of get deeper into this cube. To predict inflation cycles we have our future inflation gauge (FIG) designed to predict turning points in the inflation cycle. In 1994 and again in 1999, we predicted an upturn in the inflation cycle, an upturn in inflation pressures, and the Fed, as we know, had raised interest rates quite a bit. We also have a leading home price index, which is one aspect of inflation that a lot of people are interested in at the moment. When is the bubble going to burst, right? Well, this is designed to answer that question.
Then we have the employment side, both overall employment as well as non-manufacturing and manufacturing separately. Then we have the economic growth cycle split up into foreign trade and the domestic area. Also, we break up the domestic economy into the major sectors, you know, manufacturing, construction, services, within that financial and non-financial. And then we have foreign trade where we have leading indexes of exports, imports, and the trade balance, which are really based on a combination of domestic and foreign indicators and exchange rates and so on. We have such cubes essentially for almost 20 economies, including the G-7, China, India, and so on. These essentially interact through the foreign trade dimension for the most part.
We allow for loose linkages of cycles within and across economies. And this is interesting because this looser monitorist framework allows us to see some things fairly early that rigid and complicated models don't see until much later. For instance, employment and growth will typically move in sync but not always. So in our framework it's okay to have a jobless recovery. We predicted the jobless recovery we have had, you know, in the last few years. It's also possible to have non-inflationary growth.
Let's take an example from the 1990s. In the standard Taylor-rule framework the Fed should raise interest rates when there is an output gap between actual GDP and potential GDP because that signals future inflation. In the late 1990s the output gap soared—this is using the potential GDP estimates from the Congressional Budget Office. What did the Fed do? Not much. Fed funds rates stayed pretty low. What happened to inflation? Well, the Taylor-rule framework would have said inflation should soar. It didn't.
We did some further analysis on this over the Greenspan term, 1987 to 2001. What we found is that, contrary to what you might expect, these three variables—the output gap as a measure of underlying inflationary pressures, the Fed funds rate, and actual inflation—are not co-integrated. So if you know any two of them, let's say the output gap and the Fed funds rates, you won't have much of a clue as to where inflation is headed.
Now, if you instead replace the output gap as a measure of underlying inflationary pressures with ECRI's future inflation gauge (FIG), then you find actually they move pretty much in sync - FIG, the Fed funds rate, and inflation are co-integrated. So if you know what our future inflation gauge and the Fed funds rate are doing, you have a clue about where the CPI inflation is going. So when the Fed funds rate is following and moving in sync with future inflation gauge, you'll have pretty stable and low inflation, which is what you had. When they diverge, when FIG moves up and down but the Fed funds rate doesn't, then you would have inflation moving up and down, and that's what's happened more recently. You see that at the end of 2001 the FIG bottomed. By mid-2002, inflation bottomed. Then it followed it down. By late 2003 the FIG started going up again, and by early 2004 inflation was also trending up. What's interesting right now is that you have a downtick in the FIG to a ten-month low. If that keeps going, maybe you'll see a cyclical downturn in inflation.
Within the domestic area, we have a sequence of leading indexes: the long leading turns first, followed by the weekly leading index, followed by the short leading index, followed by the coincident index. Now, in periods of declining growth in the U.S. economy, the long leading index turns down first. Then you have the weekly leading index following suit, followed by the short leading index and finally the coincident index. So you're getting increasing levels of conviction about where you are in an economic cycle. The deceleration in growth that we have had since the spring of 2004 was clearly anticipated by this approach. The latest situation right now is that most of the indexes are still going down, but there's a little up tick—it's almost invisible—in the long leading index. We shall see how it goes, but it's just meant to suggest that it's not a single leading index that we're monitoring, but we are looking at a whole sequence of events that's happening here.
3. International coverage
In terms of international breadth of coverage, we're looking at about 20 countries in various regions of the globe. We've developed cubes similar to the one for the U.S. for all the economies we monitor and are at various stages of filling them out.
Our record of forecasting includes in many ways not just forecasting recessions in the U.S. We predicted recessions in Japan and the synchronized global recession in 2001. And almost always these calls have been against the consensus.
In some places like the U.K. we had some interesting experiences. Seven years ago, the first time we were asked to give a seminar at the Bank of England. While, like today's talk, the focus was on discussing our approach, our presentation did include a chart of the U.K. long leading index, which was pointing to a major slowdown. I tried to hide that slide a bit but didn't succeed. That very morning the Bank of England had raised interest rates. So indications of a slowdown were an awkward thing to highlight at that point. Our host, who was a member of the Monetary Policy Committee, diplomatically concluded by saying: `Well, the proof is in the pudding.' Four months later, in September 1998, the Bank did indeed start slashing interest rates. Luckily, they acted in time to avoid a recession.
I do want to mention the issue of independence because it speaks to our incentive structure. Prakash Loungani's work, in explaining the failure of forecasters to predict recessions, actually makes a little bit of a reference to the question of vested interests. But ECRI is not beholden to any constituency. We have a variety of clientele, from policymakers to strategic planners in industrial companies, to money managers.
I remember back in May of 2001, shortly after the recession began, there are some data that came out on employment, and one of the major investment banks' economists immediately issued a statement, "This is recessionary data." Forty-five minutes later, he retracted that statement. I know two other investment bank economists who actually predicted the recession correctly. Both of them have lost their jobs.
So it's important to see what the incentive structure is here.
And it's not just restricted to the dog-eat-dog investment banking industry. Because there also changing intellectual fashions in academia, we realized that it's important to be independent even from academia, and we have been so for a decade. And that's really been probably the best period in terms of how our work has flourished.
Some Econometric Issues
Let me quickly mention some econometric issues that arise in this work.
• Model specification: You have highly nonlinear processes that drive economic fluctuations. It's important not to sacrifice model specification at the altar of mathematical tractability. For instance, as I mentioned, it's important that the analysis you do is not dominated by the data away from turning points. So if you have ten years of data, 120 months, most of those months are going to be not near turning points, and yet they will dominate a standard regression. Most econometric models have a hard time with asymmetries, like some indicators lead at peaks and lag at troughs. For example, for the inflation cycle, you have labor cost measures actually lag at troughs. And I recently wrote a paper on that was posted by the Bureau of Labor Statistics on their website.
• Double counting: Another issue related to model specification—In composite index construction, there are many people who try to perform dimension reduction using principal components analysis. There, of course, double-counting is not encouraged, as in the rest of econometrics where you don't want multicollinearity in regressions. What they're trying to do in principal components is say, `okay, if one variable explains most of the variance, let's move on to what explains the residual variance' and so on. But if you think of business cycles, the point is that these are diffusion processes where what you really should be monitoring is whether it's spreading. So you do want to see that duplication, you do want to see the double-counting. You do want to see whether the downturn, for instance, is spreading from industry to industry, region to region, indicator to indicator. If you do this dimension reduction kind of thing where you don't want to do double-counting, you're in trouble. That's just an example of how we really approach this differently from standard methods. The very assumptions are quite different.
• Statistical Testing: As Granger and Newbold say in their textbook on forecasting economic time series, the leading indicators are intended only to forecast the timing of turning points and not the size of the forthcoming downswing or upswing, nor to be a general indicator of the economy at times other than near turning points. Because of this, the evaluation of leading indicators by standard statistical techniques is not easy. In practical terms, what does that mean? It means we have to use a non-parametric statistical test from the class of tests called the randomization tests. The problem here is we are trying to stick just to turning points, not data points in between. So you have a sample size of seven in this case, which is awfully small for a parametric test, but a non-parametric test can handle it without making any heroic assumptions about functional form and all that. We can test the hypothesis of whether there is a lead of a certain number of months and at what confidence level. We did a test of the CLI, which is the OECD's composite leading index for the U.S., the LEI, the Conference Board's index, and ECRI's own weekly leading index for the ECRI (WLI). It turns out at a 99-percent confidence level, the LEI has a zero-month lead, CLI has at least a two-month lead, and the WLI has a three-month lead. At a 95-percent confidence level, the CLI and the LEI have five-month leads, the WLI has a seven-month lead.
So to start winding things up here, how can we summarize all of this work we have done? I think our `monitorist' approach addresses key deficiencies in standard econometric models, specifically the ability to predict turning points. It provides useful insights during periods when the economy shows unusual behavior, for example, periods of non-inflationary growth or a jobless recovery.
It's also high capacity, low demand. What do I mean by that? It's high capacity in the sense it can monitor all the major global economies simultaneously—it doesn't just pay lip service to globalization. Few forecasting firms, if any, have had that kind of capability. It's low demand in terms of this approach's tolerance for poor data quality—lack of a system of national accounts, for example, is not a problem. If there are rapidly changing structures of the economy, as in China or India, that is also something that can be handled. And, in fact, what's interesting is that we do have a Chinese leading index, and because it's not based primarily on back data-fitting, we can do it. It seems to be working fine now, even though if you look back a decade or two ago, it didn't work that well historically. But we're not trying to fit the data to make it look pretty.
Let's be very clear. Forecasting the magnitude of variables such as growth and inflation is not the strength of this approach. And very importantly, you cannot use this approach to ask `what-if' kinds of policy questions. But the point is that it complements standard forecasting approaches. It tells you when to question the forecasts of your standard models. When is it that the forecast is likely to be wrong? What is it telling you that our forecasts are diverging from standard forecasts? So that's the way in which it can be complementary. And, of course, that's why it can be a valuable aid in decision making.
In concluding, I can do no better than quote Andrew Filardo who wrote last year: "The impressive insights of Geoffrey Moore into the theory and construction of the leading index will continue to shape our understanding of business cycles well into the future."
Thank you for your time.
MR. LALL: Thank you very much, Anirvan. That was very lucid and thought-provoking. I'm pleased to welcome as a discussant Frederick Joutz, Associate Professor and Director of the Research Program at George Washington University. He contributes forecasts to the Philadelphia Fed Survey of Professional Forecasters, and forecasting economic trends is clearly one of his main research interests. Welcome, Fred.
MR. JOUTZ: Good afternoon. Thank you all very much for coming. I have to also thank Anirvan for sending me information and some papers to look at after reading the book—it was helpful. I'm going to tell you about the book and then try to relate their work to that of the IMF.
Business cycles are not dead. Whether you're an individual, a firm, or a central bank, you face risks by not trying to recognize turning points. The book is addressed to a lay person, but it's designed to give them a disciplined approach to recognizing that leading indicators can be used to predict turning points. Though we live in a world where we're bombarded with information from CNN or many Internet sites about what's happening in the financial markets, we're nevertheless always faced with this notion of the herd or conventional wisdom. And while the conventional wisdom may work sometimes, at other times you have to take a disciplined approach to what the data is telling you.
Last night, I was driving in my neighborhood, and for the first time I actually saw a house with a sign that said "Price reduced" on it. I've been watching the housing market for several years now. And I had seen houses stay on the market for much longer, you know, longer than a weekend as used to be the case. By the time The Economist tells you that housing prices might collapse, as it does in its latest issue, it may be too late to profit from it.
Leading indicators are a tool for forecasting and they're a tool to help reduce risk. What this book also suggests is not just that there are dangers from missing turning points, but there are also opportunities from catching turning points, whether it's just a simple hedge or actually taking an advantage.
Let me illustrate that point with another example. Several years ago, I was looking at producing models and forecasts for prices and quantities in a construction industry in North America. This is a very capital-intensive industry in which there are essentially six firms. These are firms that typically have to run at 95-percent-plus capacity utilization because of the nature of their production processes. One of the things I found was that every single one of the firms was adding significant amounts of capacity exactly when the peak had occurred. It was uncanny when you looked at these increases in capacity - the firms were all building new plants or adding on to old plants at the same time. Consequently, the price for their product could vary anywhere from about $50 per unit to as much as $140 per unit. These are folks that should definitely listen to what Anirvan and his co-author have to say.
About the Book
Let me begin with a snapshot of the book. There's three parts, ten chapters in all.
• The first part reminds us that business cycles do exist. It talks a bit about the historical research on leading indicators-as Anirvan mentioned ECRI follows the tradition of Mitchell and Burns and Moore-and then gives us some data about the experience of using them for the last 15 years.
• The second part explains how ECRI's turning point indicators have been developed. I think what's very good about this section is that it tells us not to believe in one indicator. You want to use multiple indicators. So they talk about leading indicators for the real sector, for inflation, and even for the international sector—obviously relevant for the work of the IMF.
• The third part provides some scenarios or advice for people that are going to try to use indicators. They have this nice `economic dashboard' where you can imagine you're driving along and you're looking at your gauges for inflation and output and employment and what's happening in the rest of the world. Well, I thought that was very cute. I think it was a good way of presenting it. But, of course, I'm an academic economist, and so what do I want to do? I want to climb under the hood and get oil all over my fingernails and get real dirty and find out what's happening.
Let me turn to some observations on their work:
• Dealing with Uncertainty: Let's start with the basics: the word `forecasts'. Think about breaking it up into two parts. "Fore" is just saying, okay, let's look ahead or out in front and so forth, and "cast" harks back to the original use of "forecast" in throwing the dice, looking at horoscopes and so on and so forth, which we're often accused of doing. There's a famous quote that anything can be forecast but not everything can be predicted. When we talk about prediction, we're really talking about sort of the laws of nature and things that we know absolutely can happen. When we get into forecasting, though, we're stuck more with probabilities. I think the leading indicator approach is some combination of the two, if you will: it's based on uncovering some `laws' of economic nature but there are always developments we can't anticipate or imagine on the basis of what we have observed in the past. This is where the forecasters really have to level with the consumers of forecasts and get them to realize that it's not just a point estimate that we're giving you. We're giving you a story about what's going to be happening, whether it's a story for a turning point or a story for how the economy is going to be proceeding. But as forecasters, we sometimes get a little bit arrogant or too confident in our ability, and we don't remind the consumers of the forecast of the potential risks or the degree of confidence we have in our forecast. Some of you may know the quote by Maxine Singer that says, "Because of the things that we don't know that we don't know, the future is largely unpredictable." You're in a dice game with Tony Soprano. You know the probability of any pair of numbers that will come face up. What you don't know and will not know is whether the dice are loaded.
• Uncovering the laws of economic nature: So how do they come up with their leading indicators? There are certain criteria that they use. That is the part that someone like myself would like to work more on and understand more. They say their approach relies on a consistency of recurring sequences of economic events over the business cycles, which they call the three P's:
o pervasiveness of what the measures are telling us;
o how pronounced the movement is of the indicators, that is, what's sort of the ratio of signal to noise. Anirvan was talking about a case where someone on Wall Street backed away from calling a recession despite the employment numbers. But the more typical case is for folks on Wall Street to overreact to numbers, not even paying attention to what that number represents relative to what's happened, let's say, over the last year to the numbers. It could be just noise and nothing more than that.
o Persistence—is the indicator telling us something over time? And is it being consistent in what it is telling us?
• Statistical tests: Now, this discipline of the three P's is absolutely essential, and they emphasize this throughout the book. But we might want to know a little bit more about the rules for a turn for a peak and a trough. Are there tradeoffs between using the rule and potential false errors that are coming across? One can talk about trying to evaluate their record a bit more formally by thinking about peak calls relative to actual peaks, trough calls relative to actual troughs. In ECRI's case, they've been apparently very, very good. We can also think about the average lead of the forecasts in terms of the number of months ahead it warns us about—Anirvan mentioned the results of some tests of this kind.
• Dispersion: Anirvan mentioned their work is complementary to other forecasts such as those from the Survey of Professional Forecasters, which is a group much like the Blue-Chip economic forecasts. They compile a group of forecasts from different modelers or investment bankers and produce various summary statistics such as the mean and the median forecast of the group. But from my perspective, the most important thing you want to get out of these is the spread around that mean or median value and how it varies over time. Typically that gives you some idea of whether the herd may be sort of spreading out.
Relevance to the work of the IMF
Let me close with a couple of points on the relevance of their work to the IMF. First, the IMF is concerned with recognizing crises before they happen and to attempt to avoid the damage from crises by predicting them. I think that the regular use of the leading indicators framework would be a valuable tool to use in complement to other tools for forecasting that are used at the Fund. Are there relationships between economic downturns and financial crises? Is there a direction of change between the two or is there feedback? Perhaps one can build on the work of Carmen Reinhart and Graciela Kaminsky and others. In some cases, there may be windows of opportunity where there is a slowdown but acting promptly on that information can ensure that it will not lead to a financial crisis. Again, this depends on how much advance notice one gets with their leading indicators and whether there are causal links with financial crises.
Second, we have about 40 percent of the planet's population currently in high-growth situations in China and India, which is a good thing. Their current growth rates, if we can believe some of the numbers, are between 5 and maybe 10 percent. The question that I would ask myself in trying to think about this: how much of this growth is due to the catching-up sort of framework (or the accumulation perspective) and how much is business cycles? And what is their long-run sustainable growth rate, and are they transitioning to that long-term growth rate now or not? So in this case we're interested in not just talking about leading indicators of the economy or the leading indicators of growth rates. We're also talking about a move to a whole new sustainable level of economic growth. And the question here, are there indicators that we could think about in terms of economic development that might signal this transition process? Now, some people say, well, we can look at the transition from low-tech manufacturing to more high-tech manufacturing and services as an indicator. There would be really interesting work, I think, for trying to develop composite leading indicators of this kind, particularly in the case of these two countries that are so important to ourselves in the next century.
Thank you very much.
MR. LALL: Thanks very much, Fred. Now I also would like to welcome Robert Lenzner, who is national editor of Forbes magazine. Prior to that he was with the Boston Globe and with Goldman Sachs. He is the author of the best-selling "The Great Getty," a biography of J. Paul Getty. Welcome, Bob.
MR. LENZNER: I'm a journalist, not an economist. A journalist needs hard and fast information. Let me start with two examples to show you how difficult it is to come by hard and fast information.
• I did a study of the market prognosticators of the top five Wall Street firms, starting at the end of 1999, which was the peak of the bull market, and through last year to see how well they did. Let me tell you: they did very badly. Not a single one of them was able to call all the turning points. Hardly anybody called the peak. Some of them advised that 2003 would be a good time. Very few advised that 2004 would be a bad time. In short, the people that are paid the most money to tell you what to do with your money did a poor job.
• Another more recent example of this, only last week, is when the Chairman of the Federal Reserve was expostulating on why short-term interest rates going up and long-term interest rates going down. Why is the yield curve flattening out? He raised four or five different reasons that were possibly the reasons for this phenomenon occurring. One of them was that it's predicting a slowdown, possibly a recession in the economy. Another one was the incredible amount of liquidity in the world, that there is all this money sitting on the sidelines and since interest rates in the United States are higher than they are any other place in the world, possibly this is another reason. I would say in concluding from his comments that he did not say concretely identify the cause for what is happening. In fact, he more or less admitted he did not know why it was happening. This is the Chairman of the Federal Reserve Board, Alan Greenspan. So this tells you just how damn tough it is to know what's going on. (Actually, I used this book's co-author about the yield curve as a predictor of inflation. He said in 1966, the curve inverted and there was no recession. And the curve actually did not invert in '56, '59, and '90, and yet we had recessions.)
So it shows you just how difficult it is to predict even when you use the models of the past. And what it means is, I think, to me, you can't use one model any longer. You have to have all these different models, and you have to look at them and weigh them incrementally. In every cycle there's something new going on, so it's not the same as it was the last time. This time, the price of oil keeps spiking up, dropping down, spiking up. You would think based on the price of oil that you're going into a period of higher inflation. But we are not doing that. What you've got is the prices of hard goods, mainly electronic products, are coming down fast. The latest price for the new Dell computer is $299, and it's down from $499.
That brings me to the book, and the firm, and the people who are doing the research. What is essential, what is useful to me as a journalist in what they do, is that they look at the nitty-gritty. They look at what's going on out there in great detail, in voluminous detail. They look at it consistently. Why the guys on Wall Street always get it wrong is because they're not doing this basic bottom-up work.
Let me illustrate. I have here in front of me ECRI's June 2005 U.S. Cyclical Outlook. I'm just going to read to you the headlines to give you a sense of extent of coverage of their indexes:
• "Near-term growth outlook dims." "The leading employment index fell."
• "Future inflation gauge fell to a 10-month low." "The leading home price index rose in May."
• "The long leading index ticked up in a lackluster fashion." "The short leading index fell to a three-and-a-half-year low."
• "The leading services index eased. The leading financial services index slipped. The leading non-financial services index fell." "The leading manufacturing index fell to a 37-month low." "And the leading construction index fell in May dulling the construction sector outlook."
That's pretty damn valuable, isn't it? Very few people assess the data in this very intense way. I mean Alan Greenspan is known for really looking hard at the numbers—industrial orders, the transportation shipments, the inventory levels, the amount of loans taken out on homes. But very, very few people are doing ECRI's kind of work where you add it bit by bit by bit to make a rational argument about what's going on.
My own feeling is that you can use this data in our journalism, investment managers and ordinary individual investors can use it too. For me, it told me that if you want to be safe you should be in two-year treasuries since there's so very little difference in the yield between two-year treasuries and 30-year treasuries. It seems to me common sense says you should be in and wait to see what happens, because you don't really know—even Alan Greenspan doesn't really know—exactly what's going on the yield curve.
What about the present state of the U.S. economy? I conclude from what ECRI is saying is that we're having a deceleration of the economy and we're going to have a deceleration of corporate profits. Whether the Federal Reserve raises interest rates again one or two times isn't really going to matter in the intermediate to long pull because we are not faced with a sharp jag in inflation.
Now—and I'm going to end with this—there's always uncertainty. We could have any one of a number of things happen. We could have something else happen in another oil-supplying nation. We could have another terrorist attack in the United States. We could have another terrible natural event like the tsunami. But short of that, I think looking at all the nitty-gritty and adding it up and coming to some conclusion is the only intelligent way to make forecasts. It may not be fail-safe, but it's about the best you can do. It's prudent, it's intense, and it's useful.
MR. LALL: Thank you very much, Bob.
I'd like to at this point open this up to the floor for any questions you may have, any burning issues that may have come up in the last hour and a half.
QUESTIONER: I'm Prakash Loungani of the IMF. Isn't there a danger that you're sowing the seeds of your own failure? As your work becomes more known, each time you say there's a danger of recession, everybody will start acting on your information and advice and what you predict will then not happen.
MR. BANERJI: Certainly in theory self-falsifying prophecy is possible. But I think we have a very long way to go before people are following us that closely. Most people will not do as we suggest. At least for the foreseeable future I don't see that risk very much. I think that risk is more likely to the extent that policymakers follow what we are looking at, because they are in a position to take preemptive steps to avert the very outcome that we may be predicting. That could ruin our record, but that would be okay with us. So I think for the time being, it is really not as much of a risk.
MR. LALL: Please, Fred.
MR. JOUTZ: To continue Anirvan's point—if we think about the leading indicators, recall that the lead time they are able to give us in terms of warning is less than a year, and typically it's, say, six months or seven months. If you're a central banker, say like Alan Greenspan, and you try to use your policy levers, then you typically realize that, well, we can perhaps try to change short-term interest rates today, but the impact as it transmits through the economy might take us anywhere from 18 months to two years. So the tool of the leading indicator, while it is extremely valuable in thinking about what you want the world to be like in the future, is still not going to give you enough lead time to be able to effectively crush the business cycle.
MR. BANERJI: Let me also address something slightly different that this brings up. A few years ago, the late Senator Moynihan was asked in a TV interview, by David Frost, I think, he was asked, "Senator, what in your opinion is our greatest accomplishment of the 20th century?" Now, there are a lot of good answers, and I'm sure you have them. What was interesting to me was he said, "In my view, our greatest accomplishment of the 20th century is that we have tamed the business cycle, at least in the industrialized countries of the West." (As an aside, you know if this question was also asked of a 120-year-old woman, the answer might be "Simple. It's the flush toilet." So there are different perspectives here.)
But my point is that Moynihan correctly was saying not that the business cycle has been eradicated, but that it's much less volatile than it used to be. But at the same time, what's interesting is that the volatility of the securities markets, let's say stock prices, relative to the volatility of economic growth, has gone up dramatically over the past century or so. And so even if the actual economic volatility is not quite as bad as it was, you might still be able to avoid or reduce some of the risk if you are operating in the financial markets or exposed to them.
MR. LALL: Thank you very much. Any other questions? Yes, back there?
QUESTIONER: I wondered for emerging markets how fully you are able to specify your economic cube?
MR. BANERJI: That's a good question. It's also related to the point that Fred made in his discussion. In the case of India and China, our cube is not fully developed yet, and that is true for several countries.
In the case of China, for example, what we are able to do at this point is predict the industrial cycle. We are not able to predict the employment cycle in China yet. But even a partial coverage of the cube can be valuable. I was talking to some Chinese officials a couple months ago who were visiting us about how important the business cycle is for China—this is also what Fred was getting at. You see, the business cycle is essentially part and parcel of any free market economy. So the more an economy gets liberalized, this is the baggage that comes with it, the economic fluctuations. Now, the other thing that we know is if you are mostly in the manufacturing sector or you're suppliers of commodities to the manufacturing sector, then you're even more vulnerable because of the inventory cycles that manufacturing naturally produces. So if you're going to become the world's factory floor, then you are going to be heavily subjected to those fluctuations. If there's a little slowdown in consumer spending, so that it's mildly cyclical at the consumer's end, as you go up the supply chain the volatility goes up a lot. So if you're talking about manufacturers who are supplying to the consumer, their cyclical volatility is much greater. And, of course, if you're talking about developing countries who are supplying, let's say, industrial commodities to the Chinas of the world, their volatility is the greatest.
So it is particularly important for the Chinas of the world and even more for the suppliers to China to mind the direction of the business cycle, because they're very much vulnerable to that. We are hoping that in terms of data over time the data coverage will get better. For example, we know that in China now they're already doing purchasing managers surveys.
In India, you know, we were somewhat instrumental in pushing some of the private institutions to start collecting consumer confidence data.
So I think it will take a lot of time, but hopefully we'll get there.
QUESTIONER: I'm a former staff member of the Asian Development Bank. In yesterday's op-ed page, Robert Samuleson said we should throw out the economic textbooks because they're not providing the answers as they had in the past. There were two or three factors. One of them, of course, was globalization, and the linkages-I guess you termed it "integration of the global economy"-in ways that we have never faced before. The other was that is that we are spending more than we're earning. Any of you that have a 28-year-old son or daughter with a credit card, you know that's the case. But it is a unique aspect that we have very little savings rate now, and this has not been faced in the past. My question is: In the event that there is a downturn, a recession, will the integration as a consequence of globalization make it more global? And will the fact that we are not as economically prudent as we used to be in the past make it more significant? Thanks.
MR. BANERJI: In early 2000 I debated on a public television show a gentleman called Stephen Weber, who had written an article in Foreign Affairs called "The End of the Business Cycle." He was arguing that this globalization that we had seen was going to damp down the business cycle because upturns in some places were going to neutralize the downturns in others. Anybody who knows something about system dynamics knows that the opposite actually tends to happen. You really have these fluctuations amplified when you have these connected systems, especially with these nonlinear systems.
So, yes, globalization is going to make things more exciting. You'll likely have more business cycles and, more importantly, more of this transmission than you had before. That's all the more reason to monitor what's happening to the business cycles in different economies. In fact, we have some rather arcane-seeming leading indexes, for example, of the global industrial sector, to monitor how the global industrial cycle is evolving.
In terms of the lack of prudence which you have mentioned, certainly that is an issue. These things, like consumer debt, determine how bad a recession is going to be when it arrives. With low savings, there's the danger that when a recession arrives, that's when the chickens will come home to roost, so to say. But what it does not affect is the timing of the advent of the next downturn. These are not predictors of downturns. These are not triggers. So, yes, it's a concern, but it's not a reason at all to believe that these tools won't work.
Now, nothing is perfect. Nothing works all the time. And Geoffrey Moore was wise to caution us to say that if you can predict a recession when it's beginning, you're doing very well as a forecaster, and so we should keep that in mind as well in terms of limitations and fallibility of all of this. But I can't resist just mentioning since you mentioned Samuelson and textbooks. The author of one of the most influential textbooks, Paul Samuelson, in a speech to the National Bureau of Economic Research in September 1970, said that you at the NBER have essentially worked your way out of a job, meaning there aren't going to be any more recessions. That was nine months after the recession began.
You know, there were three times in the last century when the term "new era" was falsely brought up: the Roaring '20s; the '60s—I have a book on my shelf from 1969 titled "Is the Business Cycle Obsolete?", which was the year the long expansion ended; and, of course, the '90s.
MR. LALL: We have time for one more question.
QUESTIONER: I'm with the Department of Commerce. As you no doubt know, a few years ago the Conference Board took over Geoffrey Moore's index of leading indicators, you know, and they're publishing that. When they took over its maintenance, they made a change in the components. They dropped the index of sensitive prices and added the yield curve to it. And you may have noticed that next month they're going to make another change where they're going to de-emphasize—at least as I interpret it, de-emphasize the yield curve. So, specifically, I was just curious as to what you thought of the changes that they had made in the index. But, more generally, how do you deal with methodological issues like that? In other words, you know, it looks like recent data indicate that certain components are doing better or worse. Should we drop them out? Should we add them in? How do you deal with that?
MR. BANERJI: Thank you for an excellent question. I'm glad you asked that. I think this is precisely where it's important to understand that if you have an approach that is data driven and based on back-fitting, this is an inherent hazard. That's why what I emphasized right up front in terms of the history of this approach, Geoffrey Moore's 1950 list, is that it is driven primarily first and foremost by a conceptual understanding of the key drivers of the economy, including sensitive materials prices. If you're going to be driven by recent performance, that is an inherent hazard, and that's why our approach and Geoffrey Moore's approach essentially was not as data-driven as some people think. The drivers of the business cycle from country to country, from time period to time period, that's remarkably durable. It may be in a particular cycle some indicator fails. That can happen. That is why the need for multiple indicators that Fred was emphasizing here.
The irony is that that particular omission of the index of sensitive materials prices was something that was opposed quite explicitly and clearly by venerable business cycle researchers like Geoffrey Moore and Victor Zarnowitz. All of these people opposed that step at the time they were on the Advisory Board and the Conference Board did not listen to them.
But, of course, the 2001 recession was a business-led recession, so that particular indicator would have been enormously important. Then once again, based on the history of the yield spread, so to say, with a mode of empirical effect, they introduced that at the time. Now it looks like a lot of people are saying, and perhaps correctly, that it's not going to--it is not as good an indicator of the economy now. So if you keep doing that, if you keep being driven by empirical, apparent empirical shifts, there is a problem.
I think there's a clear answer to that, which is figure out what are the indicators which are most based in a clear understanding of the drivers of the cycle and which are robust in terms of their performance. Once you have that, there is no reason to swing back and forth with whatever happened recently. If you have sufficiently wide coverage, that's good enough. The conclusion we have come to, after many decades of research, is it's not enough to have a single leading index. We have, as Dr. Lall pointed out here, we have over a hundred leading indexes, close to 20 for the U.S. alone. That gives you a sense for the nuances, as Bob Lenzner was pointing out. In this day and age, it's important to move beyond what is essentially the 40-year-old state of the art that's still reflected in both the Conference Board and the OECD approaches.
MR. LALL: Thank you. Regrettably, I have to call an official close to this right now, but if you have any further questions, perhaps you can come up and speak directly to the panelists. I'd like to thank Anirvan Banerji, Bob Lenzner, and Fred Joutz for helping with this event and making it a very interesting discussion. I'd also like to thank all of you for having taken the time to come to this event. We look forward to seeing you at the next IMF Book Forum.