27 thoughts on “Comparing shipping costs and industrial production as measures of world economic activity”
AS
Professor Hamilton,
You state,
“I close by noting that one appealing feature of shipping costs is that a measure like the Baltic Dry Index is actually available daily.4 I show how one can use this daily series along with the monthly CPI to construct a daily measure of the cyclical component of real shipping costs, calculated using the two-year difference approach in the second panel of Figure 2 (Hamilton 2019).5 This measure is plotted in Figure 3 and might be used for applications where a daily rather than a monthly indicator is needed. ”
Would the daily series you develop be useful to enhance the forecast of monthly industrial production, using a mixed frequency model? I remember that you use RATS software, but I am thinking of the MIDAS model provided by EViews.
James_HamiltonPost author
AS: There are lots of potential forecasting applications of the daily series that would be interesting to explore. An alternative to MIDAS using mixed frequency data is ragged-edge nowcasting.
AS
Thanks for the link about “ragged” or “jagged edge” models. The “ragged” or “jagged edge” forecast models seem a bit daunting to amateurs. I think I would need a numerical example to understand how to use such models, due to the non-synchronous nature of the data, as described in the linked paper, “One particular feature [of Nowcasting] is typically referred to as its “ragged” or “jagged edge”. It means that, since data are released in a non-synchronous manner and with different degrees of delay, the time of the last available observation differs from series to series. Another feature is that it contains mixed frequency series, in our case monthly and quarterly.” I do wish that academic papers would include an appendix with a worked model with data that interested readers could follow. I think such appendices would increase readership and knowledge among laymen.
Barkley Rosser
Shipping costs would seem to be closely tied to world trade. That is certainly correlated with world output, although probably indeed less so than is industrial production. An extreme example of possible disjuncture, admittedly out of date, is that during the Great Depression while world output declined by something like 20-25 percent, world trade declined to about a quarter of its previous high level, although the GD was characterized by an extreme trade war following adoption of the Smoot-Hawley tariff in 1930.
Michel Lepetit
Transportation is, with production, one of the key factors of markets efficiency. It has been so for centuries. When West European markets woke up in the XI-XIIIth century, transportation – and shipping costs- were at the heart of this economic revolution, the one which led to our present world. And transportation cost is mostly energy cost (human , horses ; timber floating ; sailing ships ; see my study on Charles Dupin’s 1826 work on macroeconomics and energy).
The “price signal” of the market was for a large part a transportation cost, so an energy cost signal. The main market at local, national and international scale at that time of sustained growth was cereal for daily bread. And wheat is caloric energy …
Many thanks to James for his work on energy 🙂
Under realistic conditions of uncertainty, we must qualify for the above. Before you have recovered the cost of leveling the roadbed of the railway, airplanes might make railways obsolete –or earthquakes might undo your work. In 1932 people expected and experienced zero or negative returns from most capital projects. That is why investment ceased. Even today, if the profit rate (pure interest rate plus premium to cover risk) got forced down to 10 or 12 per cent before taxes, business as a whole might be unwilling to undertake an investment level equal to desired full-employment saving. See Appendix, page 588, for more on this.
spencer
I have long used the CRB:Index of Industrial Commodity Prices
as a short run indicator of world economic activity. It does not
include energy prices and actually tends to be a very good
leading indicator of oil prices. But I found it to be a better
indicator than shipping prices.
2slugbaits
I wonder about the future usefulness of an industrial production index as OECD economies become more service oriented and less about producing physical goods. Specifically, could we see the day when NBER has to down weight industrial production as a component of its recession judgment?
2slugbaits
This is really quick and dirty, but I compacted the monthly IP data into quarterly averages and then took the log differences of the compacted IP data and real quarterly GDP. The resulting data covered 288 quarters (1947:Q2 thru 2019:Q1). I then broke the data into three different windows of 96 quarters each and checked the simple correlation between the log diff IP index and log diff real GDP. The first and earliest set showed a correlation of 0.82. The next set showed a slightly lower correlation of 0.78. The most recent set of 96 quarters showed a correlation of 0.68. This is nothing fancy, but it does suggest that over the long run the relationship between the IP index and real GDP is weakening somewhat. Still a lot of information content in the IP index, but probably less than there used to be.
James_HamiltonPost author
2slugbaits: Sounds like you’re talking about U.S. industrial production and U.S. real GDP.
2slugbaits
Yes, US only. At a global level I would expect the IP and GDP relationship might be more stable.
I agree with James Hamilton. Tariffs would seem to affect both global GDP and shipping, and cause both to decrease. But after the adjustment to the tariffs, and growth comes from domestic industries for each country as opposed to international trade, the trends will be different and industrial production will be a better indicator.
It would be useful to learn whether Professor Hamilton is in fact using RATS.
Julian
The Rage
Right, but growth will be permanently lower as the supply chain bubble breaks and permanently lower in trend. Industrial production now and then is a decent forward indicator of a decaying economic foundation. It doesn’t matter if you have tariffs or not. You see the oil bubble bust, but it wasn’t strong enough to trigger a recession. Industrial production basically shows that. fwiw, this july’s industrial production is going to be abysmal.
James_HamiltonPost author
Julian: Sure, I use RATS for simple time-series calculations like these. More complicated stuff I program myself in Matlab.
2slugbaits
I program myself in Matlab.
You have my sympathies. In my old office we never had enough site licenses to meet demand even though everyone hated MATLAB.
Barkley Rosser
But for the really complicated stuff, MICE is nice.
Steven Kopits
Having been a shipping banker, I can say that ships are a great deal like floating real estate and accordingly draw a great number of yahoo investors perfectly capable of ordering ships into a soft market. Once ships are built, they hang around for a long time, so over-investment in the industry can result in a durable dislocation of shipping rates. In a short interval, say a month or so, shipping rates probably tend to reflect economic activity. At the horizon of a year, new entering vessels can very much screw up the market.
Industrial production numbers are typically produced with a lag, so their assistance in forecasting is somewhat limited.
I think the industry has moved well past these sorts of measures. It’s as though we were debating the merits of horses as transportation tools. Today, it’s satellites and other remote sensing, more or less 24/7 and real time, with AI and machine learning. This data is now being fed into trading algorithms, and that will affect commodity prices, both because remote sensing is increasingly competent in measuring economic activity and because the algorithms themselves influence buy-side behavior.
SpaceKnow measures light as a proxy for GDP. Other similar companies are trying to do analogous measurements, ie, cars in shopping centers, road traffic, counting construction cranes, etc. I think most of these companies would be thrilled to work with a world-class econometrician.
2slugbaits
Steven Kopits All indices have a shelf-life and I’m skeptical of the longevity of some of these light-based indices. Looking at light isn’t just a measure of economic activity; it’s also a measure of waste and presumably advanced economies will tend to reduce waste as the price of energy increases. Smart warehouse operations don’t leave the lights on 24/7. Even restroom lights turn on and off automatically when someone enters and leaves…except in US Army restrooms which turn off after a short time leaving some folks literally sitting in the dark. 🙂 You’ve probably flown into O’Hare airport at night and noticed the beautiful grid pattern of sodium lights as you approach the city. It’s especially striking against the blackness of Lake Michigan. A long time ago those street lights were even brighter (and less efficient) then they are today, but someday we may have “smart lights” that only turn on when needed. And with self-driving cars there would be less need for visible light from headlights. The point is that a highly energy efficient economy might come to resemble those satellite photos of North Korea at night.
Steven Kopits
Slugs –
If you are doing remote sensing, there’s always some fudge factoring involved. What you would like are data where the coefficients change relatively slowly and, if at all possible, predictably. Light would probably meet those criteria. On the other hand, light will tend to measure extensive, rather than intensive, development. I’m not sure, for example, that New York is much brighter than it was 20 years ago, but GDP is higher. By contrast, in India or China, light is probably a pretty good proxy for economic development.
I don’t think that there’s a single silver bullet, and I didn’t mean to imply that I thought so. I did mean to say, however, that we have a lot of new emerging options and that I personally would tend to focus on technological developments rather than trying to tie down more traditional measures. Today, the hedge funds are not looking at the DBI, or not only at the BDI. They are very much using light indices and lots of other tools. But it’s still complicated — and that’s a huge opportunity for those in econometrics. We have way more data than we ever had, and we’re not sure how to use it. Schools of economics, in my opinion, should make a more concentrated effort to better understand the world their students will be facing.
James_HamiltonPost author
Steven Kopits: That’s a useful insight, and I appreciate it. One of the things we need for these applications is a long time series. For example, the industrial production index goes back to 1958. As 2slugbaits points out, we also want something that means the same thing across different years.
Moses Herzog
There is often a oversupply of ocean ships or merchant ships which are left over in slower times. I would argue that’s not terribly insightful as I hardly watch that space in the investment world and am still aware of it. It’s kind of like grocery store shelf space you can never get rid of. Or rather maybe grocery store shelf space that is very easy to dump off (i.e sell) to other grocers at basement prices when food isn’t flying off the shelves. Of course in real life most of those landlubber grocers will close never to return, similar to what the large malls are going through right now with Amazon etc. But in ocean shipping those merchant ships (shelf space) just get unloaded on to other ocean shipping players (sometimes even leased at super cheap prices, but my understanding was the leasing actually happens more during economic upturns, but leasing is another answer in downturns to unused ships that have already been paid for).
Similar to times lags in fiscal policy and debt financing for U.S.A Presidents, I’m assuming there’s huge time lag in how the supply of ships affects different years (a new merchant ship purchased in 2010 could effect shipping rates in 2025). I would assume a lot of that lag on shipping rates has already been accounted for in econometrics models.
On the other hand, I’m relatively often nearly shocked on the things I would assume are “already” accounted for in econometrics models and are not.
Steven Kopits
I appreciate that, Jim. But if you’re looking at the Baltic Dry Index, aren’t you looking for a near real time proxy for economic activity? That’s not where you’re going to find it nowadays.
Right now, remote measurement is going through a revolution, largely as a result of cheap satellites which can be deployed for as little as $300,000, including the hardware and placement into orbit. This, along with cloud computing and the rise of AI and machine learning makes it possible to sift through thousands of images to glean relevant data–and do it pretty cheaply.
As a result, we have fabulous new tools which were unimagined just five years ago. These now make it possible to track things like economic activity, including all of shipping (which you can track by vessel to destination, by the way–you don’t need to look just at indices like the BDI–see Clipper Data), construction, light emission, car traffic, port activity, mall usage, etc. etc. For example, Genscape is using flaring activity to establish oil production in countries with weak or delayed reporting systems (eg, Russia, China, much of OPEC).
To me all this opens an incredible world of econometrics, which I think will see a real boom in the coming years. Yes, we have all that data, but making sense of it is still a lot of work–much of it of the nature associated with econometrics.
If I were structuring an Econ PhD program today, particularly in econometrics, I would want to orient the students and the activities towards the opportunities that the technology allows. These are changing rapidly and likely ushering in a golden age of econometrics.
So, I would encourage you to visit the Kevin O’Brien and his team up at Orbital Insight. They are based in Mountain View, just south of SFO. I think you would find it both incredibly enlightening and exciting.
Barkley Rosser
Jim,
This is a matter of a trivial point, but I think I would like to have it really clear what is meant by “shipping costs.” My comment above assumed that this meant aggregate expenditures for shipping, which would be the obvious measure to compare with industrial production, a highly aggregated measure.
However, it occurs to me that “shipping costs’ could mean a price index for transportation. I apologize if this is a distraction, but I am one of those economists who “swings both ways,” vis a vis micro vs macro. And indeed, although this is lower in my pile of citations, I have published in urban and regional economics, where transportation costs drive everything (location, location, location).
So I worry that perhaps when you cite “shipping costs” you might be referring to an essentially micro measure of “the cost of shipping” which would be irrelevant. I apologize that I have to ask what I think is a trivial question.
James_HamiltonPost author
Barkley Rosser: The underlying raw data are rates for shipping freight.
Barkley Rosser
Thanks, Jim. That is interesting that shipping rates are so closely tied to world output.
Barkley Rosser
Jim,
So if my first post was right that expenditures on shipping (and presumably real volumes) are more volatile than world GDP, are the rates more closely correlated with world GDP than either the shipping volumes or total shipping costs?
Professor Hamilton,
You state,
“I close by noting that one appealing feature of shipping costs is that a measure like the Baltic Dry Index is actually available daily.4 I show how one can use this daily series along with the monthly CPI to construct a daily measure of the cyclical component of real shipping costs, calculated using the two-year difference approach in the second panel of Figure 2 (Hamilton 2019).5 This measure is plotted in Figure 3 and might be used for applications where a daily rather than a monthly indicator is needed. ”
Would the daily series you develop be useful to enhance the forecast of monthly industrial production, using a mixed frequency model? I remember that you use RATS software, but I am thinking of the MIDAS model provided by EViews.
AS: There are lots of potential forecasting applications of the daily series that would be interesting to explore. An alternative to MIDAS using mixed frequency data is ragged-edge nowcasting.
Thanks for the link about “ragged” or “jagged edge” models. The “ragged” or “jagged edge” forecast models seem a bit daunting to amateurs. I think I would need a numerical example to understand how to use such models, due to the non-synchronous nature of the data, as described in the linked paper, “One particular feature [of Nowcasting] is typically referred to as its “ragged” or “jagged edge”. It means that, since data are released in a non-synchronous manner and with different degrees of delay, the time of the last available observation differs from series to series. Another feature is that it contains mixed frequency series, in our case monthly and quarterly.” I do wish that academic papers would include an appendix with a worked model with data that interested readers could follow. I think such appendices would increase readership and knowledge among laymen.
Shipping costs would seem to be closely tied to world trade. That is certainly correlated with world output, although probably indeed less so than is industrial production. An extreme example of possible disjuncture, admittedly out of date, is that during the Great Depression while world output declined by something like 20-25 percent, world trade declined to about a quarter of its previous high level, although the GD was characterized by an extreme trade war following adoption of the Smoot-Hawley tariff in 1930.
Transportation is, with production, one of the key factors of markets efficiency. It has been so for centuries. When West European markets woke up in the XI-XIIIth century, transportation – and shipping costs- were at the heart of this economic revolution, the one which led to our present world. And transportation cost is mostly energy cost (human , horses ; timber floating ; sailing ships ; see my study on Charles Dupin’s 1826 work on macroeconomics and energy).
The “price signal” of the market was for a large part a transportation cost, so an energy cost signal. The main market at local, national and international scale at that time of sustained growth was cereal for daily bread. And wheat is caloric energy …
Many thanks to James for his work on energy 🙂
By the way, I published a recent post on Paul Samuelson’s view about secular stagnation and the Zero Lower Bound. Samuelson’s 1971 famous exemple is about transportation : not shipping, but mountains leveling for railways … 🙂
https://www.linkedin.com/pulse/0-interest-rates-secular-stagnation-what-did-great-p-say-lepetit/
Interesting discussion which later has this note:
Under realistic conditions of uncertainty, we must qualify for the above. Before you have recovered the cost of leveling the roadbed of the railway, airplanes might make railways obsolete –or earthquakes might undo your work. In 1932 people expected and experienced zero or negative returns from most capital projects. That is why investment ceased. Even today, if the profit rate (pure interest rate plus premium to cover risk) got forced down to 10 or 12 per cent before taxes, business as a whole might be unwilling to undertake an investment level equal to desired full-employment saving. See Appendix, page 588, for more on this.
I have long used the CRB:Index of Industrial Commodity Prices
as a short run indicator of world economic activity. It does not
include energy prices and actually tends to be a very good
leading indicator of oil prices. But I found it to be a better
indicator than shipping prices.
I wonder about the future usefulness of an industrial production index as OECD economies become more service oriented and less about producing physical goods. Specifically, could we see the day when NBER has to down weight industrial production as a component of its recession judgment?
This is really quick and dirty, but I compacted the monthly IP data into quarterly averages and then took the log differences of the compacted IP data and real quarterly GDP. The resulting data covered 288 quarters (1947:Q2 thru 2019:Q1). I then broke the data into three different windows of 96 quarters each and checked the simple correlation between the log diff IP index and log diff real GDP. The first and earliest set showed a correlation of 0.82. The next set showed a slightly lower correlation of 0.78. The most recent set of 96 quarters showed a correlation of 0.68. This is nothing fancy, but it does suggest that over the long run the relationship between the IP index and real GDP is weakening somewhat. Still a lot of information content in the IP index, but probably less than there used to be.
2slugbaits: Sounds like you’re talking about U.S. industrial production and U.S. real GDP.
Yes, US only. At a global level I would expect the IP and GDP relationship might be more stable.
https://www.youtube.com/watch?v=I6KC0PAvJZw
Dear Folks,
I agree with James Hamilton. Tariffs would seem to affect both global GDP and shipping, and cause both to decrease. But after the adjustment to the tariffs, and growth comes from domestic industries for each country as opposed to international trade, the trends will be different and industrial production will be a better indicator.
It would be useful to learn whether Professor Hamilton is in fact using RATS.
Julian
Right, but growth will be permanently lower as the supply chain bubble breaks and permanently lower in trend. Industrial production now and then is a decent forward indicator of a decaying economic foundation. It doesn’t matter if you have tariffs or not. You see the oil bubble bust, but it wasn’t strong enough to trigger a recession. Industrial production basically shows that. fwiw, this july’s industrial production is going to be abysmal.
Julian: Sure, I use RATS for simple time-series calculations like these. More complicated stuff I program myself in Matlab.
I program myself in Matlab.
You have my sympathies. In my old office we never had enough site licenses to meet demand even though everyone hated MATLAB.
But for the really complicated stuff, MICE is nice.
Having been a shipping banker, I can say that ships are a great deal like floating real estate and accordingly draw a great number of yahoo investors perfectly capable of ordering ships into a soft market. Once ships are built, they hang around for a long time, so over-investment in the industry can result in a durable dislocation of shipping rates. In a short interval, say a month or so, shipping rates probably tend to reflect economic activity. At the horizon of a year, new entering vessels can very much screw up the market.
Industrial production numbers are typically produced with a lag, so their assistance in forecasting is somewhat limited.
I think the industry has moved well past these sorts of measures. It’s as though we were debating the merits of horses as transportation tools. Today, it’s satellites and other remote sensing, more or less 24/7 and real time, with AI and machine learning. This data is now being fed into trading algorithms, and that will affect commodity prices, both because remote sensing is increasingly competent in measuring economic activity and because the algorithms themselves influence buy-side behavior.
SpaceKnow measures light as a proxy for GDP. Other similar companies are trying to do analogous measurements, ie, cars in shopping centers, road traffic, counting construction cranes, etc. I think most of these companies would be thrilled to work with a world-class econometrician.
Steven Kopits All indices have a shelf-life and I’m skeptical of the longevity of some of these light-based indices. Looking at light isn’t just a measure of economic activity; it’s also a measure of waste and presumably advanced economies will tend to reduce waste as the price of energy increases. Smart warehouse operations don’t leave the lights on 24/7. Even restroom lights turn on and off automatically when someone enters and leaves…except in US Army restrooms which turn off after a short time leaving some folks literally sitting in the dark. 🙂 You’ve probably flown into O’Hare airport at night and noticed the beautiful grid pattern of sodium lights as you approach the city. It’s especially striking against the blackness of Lake Michigan. A long time ago those street lights were even brighter (and less efficient) then they are today, but someday we may have “smart lights” that only turn on when needed. And with self-driving cars there would be less need for visible light from headlights. The point is that a highly energy efficient economy might come to resemble those satellite photos of North Korea at night.
Slugs –
If you are doing remote sensing, there’s always some fudge factoring involved. What you would like are data where the coefficients change relatively slowly and, if at all possible, predictably. Light would probably meet those criteria. On the other hand, light will tend to measure extensive, rather than intensive, development. I’m not sure, for example, that New York is much brighter than it was 20 years ago, but GDP is higher. By contrast, in India or China, light is probably a pretty good proxy for economic development.
I don’t think that there’s a single silver bullet, and I didn’t mean to imply that I thought so. I did mean to say, however, that we have a lot of new emerging options and that I personally would tend to focus on technological developments rather than trying to tie down more traditional measures. Today, the hedge funds are not looking at the DBI, or not only at the BDI. They are very much using light indices and lots of other tools. But it’s still complicated — and that’s a huge opportunity for those in econometrics. We have way more data than we ever had, and we’re not sure how to use it. Schools of economics, in my opinion, should make a more concentrated effort to better understand the world their students will be facing.
Steven Kopits: That’s a useful insight, and I appreciate it. One of the things we need for these applications is a long time series. For example, the industrial production index goes back to 1958. As 2slugbaits points out, we also want something that means the same thing across different years.
There is often a oversupply of ocean ships or merchant ships which are left over in slower times. I would argue that’s not terribly insightful as I hardly watch that space in the investment world and am still aware of it. It’s kind of like grocery store shelf space you can never get rid of. Or rather maybe grocery store shelf space that is very easy to dump off (i.e sell) to other grocers at basement prices when food isn’t flying off the shelves. Of course in real life most of those landlubber grocers will close never to return, similar to what the large malls are going through right now with Amazon etc. But in ocean shipping those merchant ships (shelf space) just get unloaded on to other ocean shipping players (sometimes even leased at super cheap prices, but my understanding was the leasing actually happens more during economic upturns, but leasing is another answer in downturns to unused ships that have already been paid for).
Similar to times lags in fiscal policy and debt financing for U.S.A Presidents, I’m assuming there’s huge time lag in how the supply of ships affects different years (a new merchant ship purchased in 2010 could effect shipping rates in 2025). I would assume a lot of that lag on shipping rates has already been accounted for in econometrics models.
On the other hand, I’m relatively often nearly shocked on the things I would assume are “already” accounted for in econometrics models and are not.
I appreciate that, Jim. But if you’re looking at the Baltic Dry Index, aren’t you looking for a near real time proxy for economic activity? That’s not where you’re going to find it nowadays.
Right now, remote measurement is going through a revolution, largely as a result of cheap satellites which can be deployed for as little as $300,000, including the hardware and placement into orbit. This, along with cloud computing and the rise of AI and machine learning makes it possible to sift through thousands of images to glean relevant data–and do it pretty cheaply.
As a result, we have fabulous new tools which were unimagined just five years ago. These now make it possible to track things like economic activity, including all of shipping (which you can track by vessel to destination, by the way–you don’t need to look just at indices like the BDI–see Clipper Data), construction, light emission, car traffic, port activity, mall usage, etc. etc. For example, Genscape is using flaring activity to establish oil production in countries with weak or delayed reporting systems (eg, Russia, China, much of OPEC).
To me all this opens an incredible world of econometrics, which I think will see a real boom in the coming years. Yes, we have all that data, but making sense of it is still a lot of work–much of it of the nature associated with econometrics.
If I were structuring an Econ PhD program today, particularly in econometrics, I would want to orient the students and the activities towards the opportunities that the technology allows. These are changing rapidly and likely ushering in a golden age of econometrics.
So, I would encourage you to visit the Kevin O’Brien and his team up at Orbital Insight. They are based in Mountain View, just south of SFO. I think you would find it both incredibly enlightening and exciting.
Jim,
This is a matter of a trivial point, but I think I would like to have it really clear what is meant by “shipping costs.” My comment above assumed that this meant aggregate expenditures for shipping, which would be the obvious measure to compare with industrial production, a highly aggregated measure.
However, it occurs to me that “shipping costs’ could mean a price index for transportation. I apologize if this is a distraction, but I am one of those economists who “swings both ways,” vis a vis micro vs macro. And indeed, although this is lower in my pile of citations, I have published in urban and regional economics, where transportation costs drive everything (location, location, location).
So I worry that perhaps when you cite “shipping costs” you might be referring to an essentially micro measure of “the cost of shipping” which would be irrelevant. I apologize that I have to ask what I think is a trivial question.
Barkley Rosser: The underlying raw data are rates for shipping freight.
Thanks, Jim. That is interesting that shipping rates are so closely tied to world output.
Jim,
So if my first post was right that expenditures on shipping (and presumably real volumes) are more volatile than world GDP, are the rates more closely correlated with world GDP than either the shipping volumes or total shipping costs?