Monday, 3 February 2014

Steve Keen Continues in the Wrong Direction

Steve Keen has a new paper out.  Some of this simply restates points he has made before, but it also includes a revision to his controversial claims about the relationship between aggregate demand and debt.

The good news is that his new formulation avoids some of the glaring accounting contradictions of his previous work.  However, I'm disappointed that he appears to have moved even closer to relying on the quantity theory of money as the basis of his explanation for the importance of debt.  His new equation (0.15 in his paper) is essentially:

                                ADt = Yt-1 + vt . ΔDt

where AD is aggregate demand, Y is income, v is the velocity of circulation and ΔD is the change in bank lending, which Keen equates to the change in the money supply.  By referencing previous period income, rather than current, he avoids some of the problems he faced before and can therefore incorporate the usual convention that AD and Y are equal in any period.

It is perhaps worth noting that for this relationship to hold, velocity needs to be constant.  Otherwise, an additional term needs to be included to reflect the impact of any changes in velocity on the existing money stock.  It is not clear whether Keen disputes this, or whether he simply assumes constant velocity for convenience.

However, to me the bigger point is that, in his attempt to avoid his earlier accounting inconsistencies, Keen has ended up in a position which is essentially just old fashioned monetarism.  I think this is a great shame. 

Part of the problem with this approach is that Keen is forced into making a qualitative distinction between bank lending and non-bank lending.  Central to his analysis is the idea that bank lending creates money and non-bank lending doesn't.

The main difficulty with this approach in the context of a modern monetary economy is the fact that this distinction is very blurred.  It would be one thing if Keen was to specify that what matters as money is the total balance of transaction accounts.  But in fact he has to define it to include all bank deposits including term deposits.  This is because he wants to rely on a close relationship between bank loan volumes and money.  Bank loan volumes are closely related to the volume of bank deposits; the relationship with transaction account balances is much more vague, as bank customers can easily switch between sight and time deposits.

But focussing on total bank deposits as a uniquely important causal factor in aggregate demand is difficult to sustain.  An important part of the story of the crisis was the growth of close substitutes for bank deposits in the form of reverse repo and ABCP with the shadow banking sector.  It is hard to argue away the impact of the lending ultimately funded by this type of asset.

All lending generates new financial assets for end investors, whether that's bank deposits, bonds, money market funds or something else.  It may well be the case that these assets are not all the same in the impact they have on the ongoing savings ratios of the non-financial private sector.  But it is a question of degree.  It is quite wrong to suggest that there is some distinction in the mechanical process by which they arise, which makes some assets count and others not.  If we believe that, we risk overlooking some of the most important implications of developments in financial markets.

The relationship between lending and aggregate demand is a rich and intricate one.  More needs to be done to shed light on it.  Steve Keen started with a good intuition, but he has taken a wrong turning and ended up with a outdated and dubious theory.


  1. The focus of his work has been based on "debt" as it shows up in the flow-of-funds accounts. I don't see anything wrong with that approach , even though it may miss other problems.

    The problem I have with his latest paper is his use of the M2 velocity factor. First , he doesn't explain how he arrived at the conclusion that that is the correct multiplier to use , and I'm giving him the benefit of the doubt that he'll use a dynamic value to account for drift in velocity. If he's trying to model GDP , it's clearly the wrong multiplier. As his chart of M2 velocity shows , it has run between ~1.5-2.0 over the last decade. Clearly we haven't been getting 1.5-2 times the change in debt converted to new nominal gdp over this period. I suppose his answer would be that part of the debt ends up in asset values , but if so , that should be in the "aggregate demand" equation . Now he mentions it only tangentially in a footnote.

    He should stick to aggregate demand = gdp , and then model asset prices separately. At this point , if you asked him what nominal gdp will be in 6 months or a year assuming $100 billion per month in debt growth , he wouldn't be able to tell you , even though it's fairly straightforward to do so , which has seemingly been his point all along.

    I think he'll be making a correction to the correction before too long. All in all , it's a very disappointing paper.


    1. Part of the problem with the M2 measure is the point I made in this post, that the link between bank debt and M2 is not driven by accounting identities. An increase in bank debt may incrementally increase M2 at the point of making that loan, but it's all part of a continual process which also includes non-banks switching between M2 and other claims on banks. Ultimately, the level of M2 is determined more by how much depositors wish to hold in that form, than it is by bank lending.

      This then leads to the more general issue that velocity is not independent of the other variables, particularly in the fluid world of financial instruments we now live in. This approach is a dead end and he would be better off starting again on this.

      I agree also that asset prices need to be modelled separately. There are important relationships between debt levels, asset prices and expenditure and this framework is ill-suited to picking them up.

  2. Yes , the choice of M2 seems almost random. The odd thing to me is that there is an easily derived multiplier that directly links debt money to gdp , and works well in real-world use. That multiplier is simply debt/gdp flipped upside down , to gdp/debt. Here's what it looks like for Keen's "private debt" metric:

    I've subtracted federal and state/local debt from total nonfinancial debt to arrive at the flavor of debt that Keen prefers ( don't ask me why he prefers it ). If you flip the numerator and denominator to plot debt/gdp , you'll see that it matches well with fig.15 in his recent paper.

    As you can see , we fell below getting a dollar of gdp from a dollar of private debt back in the '80s. Now it's about 63 cents. That's why I'm baffled by this statement in the paper :

    "Since the velocity of money comfortably exceeds unity (though it is highly variable and pro-cyclical),
    the numerical impact of the change in debt on aggregate demand is therefore larger than I have
    claimed in research prior to developing this formal proof (Keen 2014; see also Rowe 2013)."

    That's giving us a license to pile on more debt , and it's simply demonstrably false. Keen rightly criticizes other economists for getting buried in their unrealistic models , but it seems to me he has fallen into the same trap.

    I've gone thru some sample data , taking nominal gdp and debt levels across different time periods ,and this multiplier work's well. M2 velocity as the multiplier.overstates gdp massively.

    I get better results using domestic nonfinancial debt as the metric. I don't know how you can leave federal and state debt out of the picture , since they often offset periods of low private debt growth , as in the early stages of the crisis. Here's what that multiplier looks like :

    Even when the multiplier is changing rapidly , as during the 2000s debt bubble , the rate of change is slow enough that you still get good performance . Here's a closeup of 2005-2006 :

    Ideally , you would have proxy measures based on high-frequency banking data that would allow you to adjust the multiplier between quarterly flow-of-funds releases. The big finance outfits , central banks , etc. , make these kinds of extrapolations routinely.

    Another useful change I'd like to see Keen and others make is to separate financial and non-financial sectors entirely. Study nonfinancial debt as it relates to nonfinancial value-added ( i.e. the productive economy ). Then look at the financial ( parasitic ) side separately. I think an accounting of financial debt is important , but it's vastly more complicated than the nonfinancial side , and the finance sector adds meager value relative to the reported debt load. By highlighting the sector apart from the real economy , and revealing the bloat , I think we'd soon decide that we can do with a lot less of it.


    1. Thanks for those links.

      The problem is that the relationship between debt and gdp is more complex than this. For example, it's not just the level of debt that matters but also the rate of change. Where new debt is made available to credit constrained borrowers, it may facilitate spending that would not have taken place. Ignoring further multipliers, that is a one-for-one impact, but it only happens when the loan is made, not from simply having the loan outstanding.

      I've been meaning to do more posts on how debt matters.

  3. Hi Nick, I haven't read the whole paper but what worried me was this line: 'Interest payments are transferred to the equity account of the banking sector' (beginning of Section 4 - where are the page numbers?) I know that he uses non-standard terminology but surely Keen is not suggesting that banks earn seigniorage on deposits?

    1. I find Keen's book-keeping a little different to follow, but I think this is correct. (My own version of the book-keeping is here: Interest received is an addition to bank profit and will increase equity. Interest paid would reduce equity, but as he is assuming no interest paid on deposits, the interest received represents gross profits.

      Seigniorage is the principal value of the money created (less the cost of printing). In order for seigniorage to arise, the money would have to be non-interest bearing (as it is here), but also be irredeemable. Commercial banks always face a possible requirement to redeem their deposits (even if as sector they do not). They may benefit indefinitely from funding with interest-free deposits, but as long as the possibility remains that those deposits might have to be repaid, they would not be able to recognise the principal value as a profit.

    2. Oh, but the assumption that banks can create unlimited amounts of bank deposits without paying interest is, as you say below, questionable. That's Tobin's Widow's Cruse. I'll try to get a look at your version when I get a chance, thanks.

  4. I would certainly agree with you that not all lending creates money. More specifically, there seems to be little agreement on how money is CREATED!

    To me, the problem all goes to measurement. How do we measure the economy so we can learn how much MONEY we have?

    Back in the gold days, we could weigh the quantity of gold and base our money on some relationship to the weight of gold at a specific location. In the fiat world, money is printed or keyed into existence. Where in the fiat world is our base?

    Keen seems to be looking to bank loans for a fiat base.

    In my view, bank loans only increase the money supply because every person receiving the spent proceeds of the loan thinks the money is real, good, honest, government-produced-fiat-money. The fact that the proceeds of the loan may be as they think, or, may be the proceeds from a loan of another depositor's money, never troubles their thoughts. Economist, however, recognize that when we try to measure our money supply from the total value of the nation's bank deposits, we will be counting both the amount of loans outstanding and as much government issued true-fiat money as government has allowed to remain on deposit.

    There is no question in my mind that Keen's equation 0.15 is correct as an accounting identity. The velocity term is merely an adjustment factor relating the other three terms.

  5. Hi Roger, actually it's clear that equation 0.15 is NOT correct, because some debt is created to purchase existing goods which doesn't add to aggregate demand. And also because some newly-created goods are purchased with existing money balances which does add to aggregate demand. All of this is clearly stated in Basil Moore's seminal text on Endogenous Money.

    1. Hi HJC,
      It sounds to me as if you are trying to use eq. 0.15 as a flow equation. A flow equation would map smoothly from one measuring period to another. In a flow equation, you would never have a variable such as velocity that changes from period to period without explanation.

      Equation 0.15 would be correct as an accounting equation because velocity can always be found as balancing term once the data is known.

      Equation 0.15 is incomplete as flow equation because velocity can not be predicted from the known data for the reasons you state and more.

      As you point out, not all newly created debt is fully expended in the first measurement period. When-the-newly-created-debt IS expended is a question I am just beginning to explore. MMT holds that newly created money exists until destroyed by government taxation, a period obviously longer than a year. Newly created loans-create-deposits money exists until either the loan is repaid or government tax-destroys, which ever comes first.

      I have not yet found Basil Moore's text, sorry.

    2. Roger: If I understand you correctly, you agree that it's not very insightful as a flow (or behavioural?) equation, but it's okay as a accounting identity since there can always be an ex post value for velocity - just like GDP can always be a multiple of peanut sales ex post. I agree but it also means that it's of little use.

      As for your MMT statement, the CB can 'create' and 'destroy' money by open market operations etc. Also, it probably makes sense to assume that newly-created debt is expended at the time of creation. If the new debt is to pay wages and the wages are used to buy products then the circuit of creation and destruction can be quite short.

      If you have any interest in Post Keynesian economics then having a look at Basil Moore's book would be a good idea. (Of course, that's only my opinion and you should feel free to ignore it if you like!)

    3. I agree with both of you here. Equation 0.15 is correct as an accounting identity (apart from the issue of changes in velocity which I mentioned in the post), but only if you define velocity to make it correct. In which case, it is of questionable value.

    4. HJC. I think the question of "destruction of money" needs a lot more exploration. The notion of "newly-created debt is expended" bothers me. This is not the place to explore the question but maybe, sometime in the future, it would be worth a blog post.

      Thanks for mentioning Basil Moore. He has two books potentially available on Amazon but they are both very expensive as out-of-print copies. I will look further for material that references his work.

  6. Hedging is used to counter against fluctuations, something that bitcoin investors are acutely aware of.

  7. "By referencing previous period income, rather than current, he avoids some of the problems he faced before and can therefore incorporate the usual convention that AD and Y are equal in any period."

    How pleasantly surprising. It wasn't Sisyphus after all.

    I'm so staggered by this advancement that I'm afraid to investigate the rest of the equation.

    Does he ever acknowledge his mistakes and attribute his learning?

    1. I just think it's a shame he doesn't take the time to familiarise himself with G&L, as you once suggested. It would give him a much more promising framework for his ideas.

    2. Don't worry the old equation "aggregate demand = gdp + change in debt" will appear again.