Monday, July 21, 2014

Fedwire transactions and PT vs PY

Milton Friedman's alleged license plate, showing the equation of exchange

The excruciatingly large revisions that U.S. first quarter GDP growth underwent from the BEA's advance estimate (+0.1%, April 30, 2014) to its preliminary estimate (-1.0%, May 29, 2014) and then its final estimate (-2.9%, June 25m, 2014) left me scratching my head. Isn't there a more timely and accurate measure of spending in an economy?

One interesting set of data I like to follow is the Fedwire Fund Service's monthly, quarterly, and yearly statistics. Fedwire, a real time gross settlement interbank payment mechanism run by the Federal Reserve*, is probably the most important financial utility in the U.S., if not the world. Member banks initiate Fedwire payments on their own behalf or on behalf of their clients using the Fedwire common currency: Fed-issued reserves. Whenever you wire a payment to another bank in order to settle a purchase, you're using Fedwire. Since a large percentage of U.S. spending is transacted via Fedwire, why not use this transactions data as a proxy for U.S. spending?

Some might say that using Fedwire data is an old-fashioned approach to measuring spending. Irving Fisher wrote out one of the earliest versions of the equation of exchange, MV=PT, where T measures the "volume of trade" or "real expenditure" and P is the price at which this trade is conducted. Combined together, PT amounts to the sum of all exchanges in an economy. More specifically, Fisher's T included all exchanges of goods where his chosen meaning for a good was broadly defined as any sort of wealth or property. That's a pretty wide net, including everything from lettuce to publicly-traded equities to land.

Practically speaking, Fisher wrote that it was "utterly impossible to secure data for all exchanges" and therefore his statistical approximation of T was limited to the quantities of trade in 44 articles of internal commerce (including pig iron, rice, hogs, boots & shoes), 23 articles of import and 25 of export, sales of equities, railroad freight carried, and letters through the post office. This mishmash of items included everything from wholesale goods to securities to and consumption goods. Using Fedwire transactions to track total spending is very much in the spirit of Fisher, since any sort of transaction can be conducted through the interbank payments system, including financial transactions.

Nowadays we are no longer taught the Fisherian transactions version of the equation of exchange MV=PT but rather the income approach, or MV=PY. What is the difference between the two? Y is a much smaller number than T. This is because it represents GDP, or only those goods and services that are qualified as final, where "final" indicates items bought by a final user. T, on the other hand, includes not only the set of final goods and services Y but also all spending on second hand goods, stocks and bonds, existing homes, transfer payments, and more. Whereas GDP measures final goods in order to avoid double counting, T measures final and intermediate goods, thus counting the same good twice, thrice, or even more if the good changes hands more often than that.

A good illustration of the difference in size between Y and T is to chart them. The total yearly value of Fedwire transactions, which are about as good a measure of PT that we have (but by no means perfect), exceeds nominal GDP (or PY) by a factor of 40 or so, as the chart below shows. Specifically, nominal GDP came in at $17 trillion or so in 2013 whereas the total value of Fedwire transactions clocked in at $713 trillion.




So why do we focus these days on PY and not Fisher's PT? We can find some clues by progressing a little further through the history of economic thought to John Keynes (is it a travesty to omit his middle name?). In his Treatise on Money, Keynes was unimpressed with Fisher's cash transactions standard, as he referred to it, because PT failed to capture the most important human activities:
Human effort and human consumption are the ultimate matters from which alone economic transactions are capable of deriving any significant; and all other forms of expenditure only acquire importance from their having some relationship, sooner or later, to the efforts of producers or to the expenditure of consumers.
Keynes proposed to "break away from the traditional method" of tabulating the total quantity of money "irrespective of the purposes on which it was employed" and focus instead on the narrow range of trade in current consumption and investment output. Keynes's PY measure (the actual variables he chose was PO where O is current output) would be a "more powerful instrument of analysis than their predecessor, when we are considering what kind of monetary and business events will produce what kind of consequences."

And later down the line, Milton Friedman, who renewed the quantity theory tradition in the 1950s and 60s, had this to say about the shift from PT to PY:
Despite the large amount of empirical work done on the transactions equations, notably by Irving Fisher and Carl Snyder ( Fisher 1911 pp 280-318, Fisher 1919, Snyder 1934), the ambiguity of the concept of "transactions" and the "general price level", particularly those arising from the mixture of current and capital transactions—were never satisfactorily resolved. The more recent development of national income accounting has stressed income transactions rather than gross transactions and has explicitly and satisfactorily dealt with the conceptual and statistical problems of distinguishing between changes in prices and changes in quantities. As a result, the quantity theory has more recently tended to be expressed in terms of income rather than of transactions
So there are  evidently problems with PT, but what are the advantages? Assuming we use Fedwire transactions as the proxy for PT (and again, Fedwire is by no means a perfect measure of T, as I'll go on to show later) the data is immediate and unambiguous. It doesn't require hordes of government statisticians to laboriously compile, recompile, and check, but arises from the regular functioning of Fedwire payments mechanism. There are no revisions to the data after the fact. And rather than being limited to periods of time of a month or a quarter, there's no reason we couldn't see Fedwire data on a weekly, daily, or even real time level of granularity if the Fed chose to publish it.

Even Keynes granted the advantages of PT data when he wrote that the "figures are available promptly without the necessity for any special calculation." In Volume II of his Treatise, he took U.S. "bank clearings" data (presumably Fedwire data), and tried to remove those transactions arising from financial activity by excluding New York City, the nation's chief financial centre, thus arriving at a measure of final spending that came closer to PY.

What are the other advantages of PT? While PT counts second-hand and existing sales, might that not be a good thing? Nick Rowe, writing in favour of PT, once made the point that it's "not just new stuff that is harder to sell in a recession; it's old stuff too. New cars and old cars. New houses and old houses. New paintings and old paintings. New furniture and antique furniture. New machine tools and old machine tools. New land and old land." As for the inclusion of financial transactions, anyone who thinks asset price inflation or deflation is an important property of the economy (Austrians and Austrian fellow travelers no doubt) may prefer PT over PY since the latter is mute on the subject.

I'd be interested to hear in the comments the relative merits and demerits of PY and PT. Why don't the CNBC talking heads ever mention Fedwire, whereas they can spend hours debating GDP? Why target nominal GDP, or PY, when we can target PT?

For now, let's explore the Fedwire data a bit more. In the figure below I've charted the total value of Fedwire transactions (PT) for each quarter going back to 1992. I've overlaid nominal GDP (PY) on top of that and set the initial value of each to 100 for the sake of comparison.



It's evident that the relative value of Fedwire transactions has been growing faster than nominal GDP. However, the financial crisis put a far bigger dent in PT than it did PY. Only in the last two quarters has PT been able to break to new levels whereas nominal GDP surpassed its 2008 peak by the second quarter of 2010. Is the financial sector dragging down PT? Or maybe people are spending less on used goods and/or existing homes?

Fedwire data is further split into price and quantity data. Below I've plotted the number of transactions, or T, completed on Fedwire each quarter. On top of that I've overlaid real GDP, or Y. The initial value of real GDP has been set to 16.6 million, or the number of transactions completed on Fedwire in 1992.



After growing at a relatively fast rate until 2007, the number of transactions T being carried out on Fedwire continues to stagnate below peak levels. In fact, last quarter represented the lowest number of transactions since the first quarter of 2012, a decline that coincided with the atrocious first quarter GDP numbers.

Finally, below I've plotted the average value of Fedwire transfer by quarter. On top of that I've overlaid the GDP deflator. To make comparison easier, I've taken the liberty of setting the initial value of the deflator at the 1992 opening value for Fedwire transaction size.



As the chart shows, the average size of Fedwire transfers really took off in 2007, peaked in late 2008 then stagnated until 2013, and has since re-accelerated upwards. In fact, we can attribute the entire rise in the quarterly value of transactions on Fedwire (the second chart) to the growth in transaction size, not the quantity of transactions. Fedwire data is telling us that inflation of the PT sort has finally reemerged.

A few technical notes on the Fedwire data before signing off. As I've already mentioned, Fedwire provides a less-than complete measure of PT. To begin with, it doesn't include cash transactions (GDP does, or at least those that have been reported). This gap arises for the obvious reason that cash transactions aren't conducted over Fedwire. Nor do cheque transactions appear on Fedwire, or at least they do so only indirectly. Check payments are netted against each other and canceled, with only the final amounts owed being settled between banks via Fedwire, these settlements representing just a tiny fraction of the total value of payments that have been conducted by check over any period of time.

The same goes for securities transactions. Fedwire data underestimates the true amount of financial transactions because trades are usually netted against each other by an exchange's clearing house prior to final settlement via Fedwire. The transfer of reserves that enables the system to settle represents a small percent of the total value of trades that have actually occurred.

Another limitation is that Fedwire data doesn't include wire payments that occur on competing payment systems. Fedwire isn't a monopoly, after all, and competes with CHIPS. I believe that once all CHIPS payments have been cleared, final settlement occurs via a transfer of reserves on Fedwire, but this final transfer is a fraction of the size of total CHIPS payments. And finally, payments that occur between customers of the same bank are not represented in the Fedwire data. This is because these sorts of payments can be conducted by a transfer of book entries on the bank's own balance sheet rather than requiring a transfer of reserves.

I'm sure I'm missing other reasons for why Fedwire data undershoots PT, feel free to point them out in the comments. Do Fedwire's limitations cripple its value as an indicator PT? I think there's still some value in looking at these numbers, as long as we're aware of how they might come up short.

Some links:
1. Canadian Large Value Transfer System Data, the Canadian equivalent to Fedwire
2. A paper exploring UK CHAPS data,the British equivalent to Fedwire: Income and Transactions Velocities in the UK

* 'Real time' means that payments are immediate and not subject to delay, while 'gross settlement' indicates that payments are not grouped together for processing but submitted individually upon being entered. Fedwire gets its name from the beginning of the last century, when payments were carried out over the wires, or the telegraph system. 

22 comments:

  1. very interesting post - this sounds similar to the benefits and problems associated with Gross Output - which the BEA just started releasing

    Some questions

    What about federal funds trading?

    One point to add to your CHIPS discussion - with megabanks what happens if a bunch of transactions take place and everyone banks at JPM Chase? Then its all internal to JPM and there's nothing on fedwire or any other settlement system for that matter right?

    ReplyDelete
    Replies
    1. "What about federal funds trading?"

      Good point. Fedwire used to handle billions of dollars of fed funds trading but that came to a halt early in the credit crisis. Fedwire volumes have surely been affected by this. I would have expected most of the damage to be in the data by 2010 or 2011. That the transactions data continues to lag would seem to be due to something other than stagnant fed funds markets.

      "One point to add to your CHIPS discussion - with megabanks what happens if a bunch of transactions take place and everyone banks at JPM Chase? Then its all internal to JPM and there's nothing on fedwire or any other settlement system for that matter right?"

      You're right, those transactions wouldn't be captured. So ongoing bank consolidation would have the effect of reducing Fedwire transactions but it would be a mistake to take this reduction as an indicator of reduced economy-wide transactions.

      Delete
  2. Great post, I'd been wondering what PT vs. PY roughly looks like.

    A bit of a tangent:

    Fedwire is probably the most important financial utility in the U.S., if not the world. Member banks initiate Fedwire payments on their own behalf or on behalf of their clients using the Fedwire common currency: Fed-issued reserves.

    I don't remember anyone specifically mentioning this in the "why does fiat money have value?" discussions, but I had a thought that fiat money (in the form of central bank reserves) gets a lot of its value in the same way that XRP does in Ripple: as a ticket to participate in the only payment-processing system in town. Even though Fedwire is not a monopoly, ultimate settlement in CHIPS occurs over Fedwire, so at the end of the day everyone needs Fed-issued reserves to stay in business.

    No player in the financial industry has any incentive at all to establish an alternative payments system using a different settlement asset (e.g. gold or stocks) due to: 1) the huge inertia of network effects; 2) tax disadvantages (e.g. capital gains taxes on gold & financial assets); and 3) the threat of losing explicit (FDIC) and implicit (TBTF) subsidies from the central bank upon establishing a rival, non-central bank reserves based, payment system.

    So fiat money wins by default in a landslide. Rounded out with tax acceptability and Mike Friemuth's debt-chartalism, the positive value of fiat money doesn't seem like a huge mystery (even w/o central bank asset "backing"). The upshot is that the growth of alternative payment systems like Ripple or the defunct E-gold constitute the biggest threat to the dollar.

    ReplyDelete
    Replies
    1. John, that's an interesting theory.

      But what happens when a monopoly like the Fed or Ripple issues these instruments to a degree that far exceeds any demand for them as participation tickets? The price of the marginal ticket should plunge to 0 since the issuing institution has created too many units. If we look at the Fed, this is exactly what is going on right now. The quantity of reserves held by banks far exceeds the amount necessary to settle Fedwire transactions. Why are banks holding these reserves despite the fact that the marginal unit of reserves is useless as a settlement medium? In other words, why isn't the price level plunging?

      I'd argue that since the Fed holds assets, it can afford to pay interest on reserves, and this is encouraging banks to hold on to them. If the Fed didn't hold assets it wouldn't be able to pay interest, nor would it be able to repurchase reserves in order to support their value, and the price level would have fallen much further.

      I should say that I agree with you in part. In general I think that the monopoly nature of Fedwire is what gives reserves their high liquidity premium (in normal times), higher than what those same instruments would earn in a free market.

      Delete
    2. why isn't the price level plunging?

      Shouldn't this read, "Why isn't the price level exploding?" Normally, as the convenience yield of the marginal ticket declines to zero, banks should be eager to rid themselves of reserves and bid up the price of other assets, correct?

      [ But if the "participation ticket" theory is right, might it be possible that they can't rid themselves of the tickets because there are no willing takers, since everyone's already stuffed to the gills with reserves? ]

      If the Fed didn't hold assets it wouldn't be able to pay interest

      I don't see why this isn't true. Won't a few keystrokes suffice?

      Although it might be bumpier than using OMOs, in theory couldn't a CB conduct monetary policy purely by manipulating reserve ratios (i.e. altering the convenience yield of reserves) and/or IOR (pecuniary returns)? It seems that no assets would be required.

      Delete
    3. Why are banks holding these reserves despite the fact that the marginal unit of reserves is useless as a settlement medium?

      If the Fed didn't hold assets it wouldn't be able to ... repurchase reserves in order to support their value

      Ok, I think I get what you're saying here. Banks are holding onto reserves because they expect that sometime in the future, the Fed will conduct open market sales allowing them to convert those reserves into other financial assets (Treasuries, MBS). So in that sense these assets "back" reserves. [ Btw, I read an old post of yours on front-running and QE. Is it possible that banks are sitting on reserves waiting to do the opposite, i.e. stand pat until they can scoop up assets at cheap prices since the big, dumb trader will likely unload its holdings in a predictable manner? ]

      I believe I saw on Glasner's blog that John Cochrane and others have come around somewhat to accepting the idea of a permanently enlarged Fed balance sheet, and many commenters at Sumner's blog like to point out that an exit strategy is unnecessary. If an exit strategy doesn't seem to be forthcoming over the next decade or so, I wonder how this will affect bank behavior w.r.t. holding reserves.

      the price level would have fallen much further.

      Again, I'm confused. Assuming that the CB doesn't have assets, and reserves play the role that gold did in the gold standard (settlement medium), then shouldn't a fall in the value of reserves (due to increased supply) result in an increase in the price level? As Glasner recently pointed out, the opposite case--a rise in the value of gold--led to a falling price level which culminated in the Great Depression.

      Delete
    4. "Shouldn't this read, "Why isn't the price level exploding?" "

      Yes, you're right. More later.

      Delete
    5. Good comments, John. Plenty to digest:

      "I don't see why this isn't true. Won't a few keystrokes suffice?"

      Yes, that is one way to pay interest. However, it won't work for long. Imagine a new private bank opens its doors for business. It pays depositors interest by crediting their accounts via keystroke but absconds with the assets. When depositors start to question the state of the bank, it increases its keystroke rate, thus appeasing them. If the bank could continue doing this, we'd call it a stable ponzi scheme; it is earnings something for nothing. Actual banks don't operate this way. The interest they pay depositors comes from the earnings that its assets generate. I don't think the nationalization of a bank changes things.

      "Although it might be bumpier than using OMOs, in theory couldn't a CB conduct monetary policy purely by manipulating reserve ratios (i.e. altering the convenience yield of reserves) and/or IOR (pecuniary returns)? It seems that no assets would be required."

      Yes, I agree that it could do this. And I agree that reserve requirements would be a sufficient (though not necessary) condition to generate a positive demand for central bank liabilities without necessitating assets. In this case we'd have something like XRP, where the necessity of having some tickets on hand drives their demand. ( I believe you need a reserve of XRP to conduct a transaction?)

      But there are many central banks that don't have reserve requirements (ie the Bank of Canada) yet their liabilities continue to earn a positive value. And secondly, as I've pointed out above, the Fed has currently issued far more balances than are required to meet reserve requirements. This means that, assuming that the only driver of the demand for Fed balances was to meet margin requirements, their marginal value should be $0. That Fed reserves aren't worth $0 (we have even had a minor deflation) implies that something (like the assets-theory-of-money) must be working behind the scenes.

      To further elaborate: I agree that reserve requirements are capable of driving a positive value of fiat money, and that modifications to the level of these requirements could thereby alter their value. But I don't think that reserve requirements are the only way to get the ball rolling and in the case of central bank money, something else is driving the process.

      Delete
    6. "But since these assets themselves are promises to pay money, they don't seem like good assets to back the money supply, as Mike Sproul mentioned before."

      Yes, that's definitely an issue if you buy the backing theory. Bond backing should make the price level highly unstable. (I've always thought that a good criticism of the theory is that the price level appears to be far more stable in practice than it should be in theory.)

      "I remember Glasner being incredulous (he called it "preposterous") when presented with the idea that FRNs entitle the bearer to any assets of the US govt (and he didn't reply when you linked to the fine print, which I think was a bit shabby of him). "

      Yes, I had been hoping to get a response from him as he usually makes an effort to address all comments left on his blog. (The post that John is referring to is this one.) Either David was too busy to respond, or maybe I managed to slide one by him that he couldn't parry.

      "But let's look at it another way--if even Glasner was unaware of this bit of legalese, you can bet that pretty much everyone else in the world is, too."

      That's a fair point. My only response is that the number of people who drive the reserve valuation process is small and anonymous, a small cadre of dealers that does not include you, me, or David Glasner (well, it might include you, but I know it doesn't include either myself or David). These traders interact with a central bank's open market operations desk on a daily basis. They expect to submit assets in order to get balances, and return those balances at the end of the day to get their assets back. While the precise legalese may not be widely understood, I'd argue that a "bankers" or "backing" mentality could tacitly emerge via these constant quid pro quo transaction.

      "On the other hand, the necessity of having reserves for settling and making payments is quite obvious to everyone in the financial industry."

      I can buy that. But when the system is flooded with reserves beyond all necessity for settling, making payments, and meeting reserve requirements, why are they still valuable?

      Delete
    7. Good question on the value of reserves. I'll post a few more thoughts next week.

      Delete
  3. The problem with this is it seems to depend very much on institutional structure. It's a similar issue to what you get if you try to interpret the level of financial sector debt.

    Traditional banks are a one-stop shop with loans on the asset side and money instruments on the liability side. Shadow banking splits the various elements of credit transformation, so between end investor and end borrower, there are normally several entities in a row. The level of debt looks higher, but it's really just counting the same debt several times. The same thing happens with the volume of transactions.

    The same thing applies with production of goods, of course. If a retailer acquires its supplier, the level of transactions falls, but nothing much has really changed. But I think it has a much bigger impact in the finance sector.

    ReplyDelete
    Replies
    1. "The level of debt looks higher, but it's really just counting the same debt several times. The same thing happens with the volume of transactions."

      I kind of see where you're going with your debt example. I would just say that PT isn't about counting the same transaction twice. The idea is to count every transaction once (over a period of x). If the same good appears in a ten transactions during x, then we add 10 to our running tally of transactions. Double counting items is not a sin with PT, double counting transactions is.

      "If a retailer acquires its supplier, the level of transactions falls, but nothing much has really changed."

      Not sure what you mean by 'really changed'. If one thinks in terms of PT, then a retailer buying its supplier (and thus reducing transactions) means that all other things staying the same the price level will have to rise. With PY nothing changes.

      Delete
  4. I strongly disagree with that quote from Keynes. There is no reason that trade in labour, or in newly-produced goods, should be more (or less) welfare-improving than trade in anything. Recessions disrupt trade, and so reduce welfare. Why we should restrict our attention to disruptions of trade in labour and newly-produced goods is beyond me.

    ReplyDelete
    Replies
    1. I know you've come out in favor of NGDP targeting, which is a PY measure. Would you prefer a transactions based target, PT targeting rather than PY targeting?

      Delete
    2. JP: Dunno. Probably not. Mergers of firms with their suppliers, or waves of financial transactions, might cause monetary instability.

      Delete
    3. Wouldn't waves of used-good transactions cause monetary instability too? Waves of buying and selling existing homes, for instance?

      Recessions disrupt trade and trade reduces welfare, so a disruption of financial trade reduces welfare. Why should we focus only on trade in labour, newly produced goods, and old goods when a disruption of trade in financial products is also welfare reducing?

      Delete
  5. This is something I think about often, as I think many people have lost sight of the history of the equation of exchange when they discuss it. I once heard Ray Dalio say in an interview "velocity is a made up thing," and the interviewer never took him to task on the statement, but I think he was touching on this issue of a lack of clarity, as V depends not only on what we mean by M, but also by how broadly/narrowly we define T/Y.

    One thing that I think is very interesting is that T has some liquidity aspects to it. If assets start cycling more frequently and at higher volume, T should rise proportionately even if nothing has been created or prices aren't inflating. I don't think of this as a bug but as a feature. T simply encompasses more things, one of which is liquidity. The measure is also affected by the level of vertical and horizontal integration, as many people have already noted.

    Considering how much information is bundled in these single measures, I think it is important to have a wide array of them from which to look at instead of the current approach which enshrines GDP in a monolithic aura. One which comes in between T and Y I think would be very important. Measures like Gross Output get close, but I believe that measuring goods-in-process could be very enlightening. This could perhaps be proxied by looking at something like revenues rather than income. It is like T except without including transactions in assets. Let me know your thoughts.

    Also, here are some of George Selgin's thoughts from a post I remembered from a while back on the same issue.

    http://www.freebanking.org/2012/10/01/intermediate-spending-booms/

    ReplyDelete
    Replies
    1. "I don't think of this as a bug but as a feature."

      Yes, I think that's a great way to put it.

      "...here are some of George Selgin's thoughts from a post..."

      I remember that post well. You'll see my old comment on it at the bottom. I had originally meant to link to it here but never got around to it.

      "Considering how much information is bundled in these single measures, I think it is important to have a wide array of them from which to look at instead of the current approach which enshrines GDP in a monolithic aura. One which comes in between T and Y I think would be very important. Measures like Gross Output get close, but I believe that measuring goods-in-process could be very enlightening. This could perhaps be proxied by looking at something like revenues rather than income. It is like T except without including transactions in assets. Let me know your thoughts."

      I don't know much about gross output (Will mentioned it too, see above) as it only recently began to be released on a quarterly basis. If I remember correctly, it measures all final goods and services and also all intermediate goods? If so, wouldn't gross output less GDP give you goods in process?

      Delete
  6. I think that's basically right, but if I remember correctly I think Gross Output excludes transportation costs and some other downstream things, which I would be interested in as part of the production process.

    ReplyDelete
  7. "I think there's still some value in looking at these numbers, as long as we're aware of how they might come up short."

    Short of the true long really - which points to how dangerous it is to build any analytical framework around such a measure.

    It's interesting information.

    But needs to be decomposed in context.

    GDP is conceptually closed and pure as a measure of income.

    This asset churning stuff is not. Everybody is skimming a bit off the top - or trying to.

    Sure its useful. It's all useful. But there is no conceptual centricity to this sort of measurement.

    Empirical velocity is an inherently reckless concept.

    ReplyDelete
    Replies
    1. Hi JKH.

      "GDP is conceptually closed and pure as a measure of income."

      Well they're both identities and equally 'closed and pure'. I do agree that Fedwire comes up short in measuring PT, probably more short than GDP does in measuring PY. There are plenty of black market and misreported transactions that don't get included in GDP.

      Also, the T we're talking about isn't velocity, its a different concept. An increase in MV due to a rise in V can have no effect on T, as long as prices take all the impact.

      Delete
    2. Good points. It was a sloppy comment I guess.

      Pure in the sense that the addition of specific value to GDP happens in an identifiable, discrete way. There is a well-defined aggregate netting concept to arrive at the result.

      By comparison, there is no such aggregate netting process for much transactional volume. For example, stock market, foreign exchange, and two way gross international capital flows can blow up volume without a similarly well-defined net value effect.

      Delete