Mortgage Finance Association of Australia Talk
Mar20
on March 20, 2011
at 7:21 am
I spoke at a MFAA Professional Development Day, following a speaker who pointed out that most decisions are made by the emotional components of our brains–hence some of my references to using the CEO segment of your brain instead.
Click here for the PowerPoint slides.
I disagree AK, on the basis of the formula that vk subsequently agreed to: AD(t) = GDP(t-tau) + dD(t)/dt. This is something you will find in Schumpeter, Minsky and Marx. I believe that the Godley et al analysis on this falsely brings in a conservation law that does not apply in our credit driven economy.
Schumpeter put it best:
THE fundamental notion that the essence of economic de-velopment consists in a different employment of existing services of labor and land leads us to the statement that the carrying out of new combinations takes place through the withdrawal of services of labor and land from their previous employments… this again leads us to two heresies: first to the heresy that money, and then to the second heresy that also other means of payment, perform an essential function, hence that processes in terms of means of payment are not merely reflexes of processes in terms of goods. In every possible strain, with rare unanimity, even with impatience and moral and intellectual indignation, a very long line of theorists have assured us of the opposite…
From this it follows, therefore, that in real life total credit must be greater than it could be if there were only fully covered credit. The credit structure projects not only beyond the existing gold basis, but also beyond the existing commodity basis. (Schumpeter 1934, pp. 95, 101; emphasis added)
also…
We must not confuse difference equations and discrete time modelling with differential equations and continuous time domain. Therefore aggregate_expenditure(t) must always be equal to aggregate_income(t) (continuous time) or aggregate_expenditure(t_n) = aggregate_income(t_n) (discrete time)
It is not that AD(t) = GDP (t minus lag) + dD(t)/dt
t – tau is used in difference equations and it is t_n-1
Nyquist-Shannon sampling theorem binds together synchronous-digital (discrete) and analog (continuous time) signal processing approaches. As long as sampling frequency is high enough both approaches are equivalent. W. Godley was using discrete time domain but this doesn’t change anything as his models can easily be transposed to the continuous time domain.
If we have positive dD(t)/dt (new mortgage loans creation) this will be seen in NIPA as a temporary increase in both I (investment) and S (savings) as (by definition)
Y = C + I + G + (X – M) = C + S + T
Nobody will notice what’s wrong. Everyone is happy as the economy is humming, investment and savings have risen, etc… In fact C will also rise (due to the spending multiplier – see below).
It is a bit more tricky to see the impact of rising consumer credit as it may manifest itself as falling S (compensated by rising T-G leading to a budget surplus or/and falling X-M leading to trade and current account deficits) or may not be visible at all in NIPA – except for an unexpected growth of C
Let’s split households into H1 (savers) and H2 (spenders) and assume that G=T and X=M.
H2 draw new credit from the banking sector at a rate dL/dt = S2 and their contribution to saving is (-S2) They spend this amount of money on plasma TVs, vodka, gambling, overeating, etc, what makes them instantly happy. This extra spending increases C by S2 and the GDP increases (all things equal and assuming spare productive capacities) by S2 in the first period as well. The mechanism is self-feeding as in the second period the economy grows further. The length of the period (the time constant) is determined by the average production-sale period which is closely related to the money velocity in the productive economy (provided that we defined the money correctly). Can the economy grow forever because of the extra loading? Assuming linear taxation and zero marginal propensity to import it is up to households H1 to remove the additional flow. Let’s assume that they save 20% of their extra income (marginal propensity to save) which is equal to 0.5 of GDP. So they can remove 0.1 of the delta. Let’s assume zero marginal saving propensity of H2 The new equilibrium will be reached at S1 = S2 but knowing that delta S1 = 0.1*delta Y
we will get delta Y = 10* dL/dt
In this model as long as no independent investment decisions are made, I=S
Obviously the stimulatory effect of consumer debt loading on the economy may be throttled due to taxation, changes in G (“automatic stabilisers”) and non-zero marginal propensity to import. However I can guarantee that no neoclassical economists will spot the real cause of the mysterious great moderation… that is the growth of the cancer of debt and deposits in the banking sector gobbling the whole economy (contributing to over 10% of GDP in Australia now – not by new credit but by income redistribution). We also need to consider secondary effects of the investment decisions where the accelerator kicks in. This is the real dynamics which is driven by credit binge and when debt loading goes into reverse, the multiplier also goes into reverse and the negative effects of the crisis can only be moderated by increased government deficits as long as this is politically palatable.
In the model outlined above the quantity of money (credit) does not affect the GDP as the causation doesn’t go that way. Monetarism is incorrect and the lag is not constant. There is no voltage in economics and the value of GDP(2011) (flow) mainly depends on the value of GDP(2010) (another flow) and some other flows not on the stock of M1, M2 or M3 however the stock of debt (and savings) does influence the flows due to interests accrued on loans (again this is invisible in the highly aggregated GDP statistics).
In this context to analyse the credit impulse one has to disaggregate the changes in debt into the component related to repaying the old long-term loans – deleveraging – and taking new short-term productive loans (increasing the size of the revolving fund used to finance production). If the half-decay time of loans (“money velocity”) is shortened, there can be an increase in the economic activity even in a pure credit economy when the overall level of debt decreases.
I am fully aware that the analysis presented above is extremely oversimplified but my goal was to show how to tackle the mystery the impact of debt supercharging on the economy rather than to build a dynamic model since what’s really interesting is the behaviour of the speculators (affected by the changes in the value of assets), the impact on investment in production (the accelerator effect), changes in the inflation rate and unemployment, secondary effects caused by the changes in the interest rate etc… It will take years to get everything right but as long as the basic model is not sound I can guarantee that the model simply won’t work.
Steve,
Questioning accounting principles will not lead us anywhere. I am coming from a perspective of a guy who first writes down numbers on a piece of paper and then tries to draw any conclusions. This is the only working approach in engineering even if analysis and design involve top-down stages – inductive thinking is the key to get everything right. So I will first examine balance sheets of a firm, worker and a bank, then aggregate them and add the treasury and the central bank. Only at this stage I can start examining the nature of relationships between parts of the system and in the end – the nature of money.
W. Godley’s models are consistent with the reality, please find a single hole in his maths or any inconsistency between his models and the reality. If a model is inconsistent with the accounting identities then it is not a model of the financial economy. Do I need to quote everything here again? If after a productive cycle all the commodities are consumed (or invested as productive capital) by the workers, capitalists and bankers the initial loan can be repaid and credit/money destruction can occur – but this obviously doesn’t make sense in the real life so the next productive cycle starts immediately. It is not the quantity of money lent what determines the outcome. It is the amount of real assets which can be used as a lien to secure a loan and the expected revenue, costs and profits. The only relevant lag is determined by the production period length what is not constant because sometimes selling production takes longer than anticipated (and varies depending on the industry). After aggregating multiple firms the quantity of money (the size of the revolving fund of credit) has its upper limit because it is determined by the value of the real capital goods (assets) but it is also an artefact of the business activity – hence there is another fundamental error in the causation of the models stating that the amount of credit determines the production level. Again – it is the value of the productive (real) capital and the demand. The demand depends on the wages, changes in consumer credit and investment decisions regarding the next cycle. That’s why QE doesn’t change anything – banks don’t lend despite rolling in liquidity and companies have a lot of cash sitting on their accounts and do nothing until there is enough anticipated demand to restart the production of goods.
I didn’t read the whole Capital but from what I saw Marx also got the basic production circuit right – what he messed up was his value theory because as a philosopher he wanted to create the ultimate theory of everything. Hoarding money is not a goal of the capitalists (except for very special circumstances), they are after increasing profits and building wealth (retaining some of the profits). They want to be rich – own the biggest corporations, have huge mansions, planes, yachts, land, other real assets, maybe some government debt securities. Only a certain small percentage of their assets are deposits in the banking system withdrawn from the circulation.
The fundamental mistake is in my opinion related to assuming that RETAINED profits in the basic monetary circuit are monetary – they are mostly in real assets. I have already sent a link to the paper explaining that. This could explain the fundamental error leading to a leak in the models. The total value of savings and bank equity in a pure credit economy must be always be equal to the aggregate debt and there is no mysterious way of accumulating profits based on bank vaults, notes or whatever. It is not the act of printing bank notes what creates money (tokens), it is the creation of transferable IOUs which can be extinguished in a similar way they were create – like old debt tallies. Again this is so obvious to anyone who ever worked with any accounting database systems. In the end the asset and liabilities sides on all the balance sheets must offset themselves and the only way to get there is to acknowledge that there is a kind of “conservation” law (I’m not sure what you mean by it but it sounds about right to me). Only the government can create money out of thin air.
Whatever can be added can also be subtracted. Every transaction which can be done can also be undone. The only exceptions to symmetry are the flow of time which goes in one direction and the interest accrual (it is hard to imagine people voluntarily allowing for negative interest rates).
If we question these basic accounting rules and want to speculate what money is what is not, I am afraid we will not be able to build a viable theory any better than these invented by Milton Friedman. It may be enough to get a Nobel price in economics but may not be enough to explain what’s going on.
Thanks for the replies Steve and others. I think i got most of it, but i got a little lost in the end.
Hi Steve,
Great talk as always.
Just one question. Do you think there is a big difference in how everything plays out given that the RE bubble in Australia consisted mainly of bidding up existing house prices, compared to the US and say Ireland, which had a lot more additional building? I’d be interested in your thoughts on this.
Cheers,
Ben
Thanks Hatless,
And yes I do think that’s a significant difference in how things will play out here–though not as the spruikers see it of course.
Their argument has been that since we didn’t build as many houses here as in the US (true), there won’t be the same overhang of unsold new properties depressing prices (true). Therefore our prices won’t fall (false).
What they’re omitting from their thinking is that, given that Australians borrowed more money to gamble on housing than even the Americans did, our bubble was more of a purely speculative one than theirs was. They at least did some “investment”, even if it was inappropriate to needs in the medium term. We did far less, so far more of our money went to gambling on house prices than increasing the quantity of housing.
This is one reason our price bubble was more extreme than theirs, and therefore potentially has much further to fall.
It also will spread the pain of a price fall more broadly. Whereas we will have less losers among property developers, we will have more losers amongst those who bought an existing property for capital gain as a retirement investment. Something like 30% of market demand came from “mum and dad investors” at the peak of the bubble. If a significant proportion of them think that the longer they hold a property, the less they’ll have for retirement, then–with a large lag–they could switch from the buy side to the sell side.
Love how stuff comes out in the wash when things are tight…
http://www.guardian.co.uk/business/2011/mar/21/ons-inflation-cpi-rpi?CMP=twt_fd
Wont be long b4 it happen here either.
AK
“Nyquist-Shannon sampling theorem binds together synchronous-digital (discrete) and analog (continuous time) signal processing approaches. ”
The sampling theorem does no such binding. It defines the minimum sampling rate that will enable the digital representation to be used to reconstruct the original signal – data stream. This is achieved as it avoids the effect of aliasing. In any case most economic data is sampled at a far too low rate to meet the Nyquist-Shannon requirement. Difference equations are part of the discipline of discrete mathematics and are not valid in the description of a dynamic system. They are not used for this purpose by engineers. Differential equations are the only valid approach.
The debt supercharging is best illustrated by the positive feedback loop of a dynamic system, too much loop gain and you don’t need any input (deposits) to get an output (credit). In days gone by before the frequency shift method was used to avoid the problem everyone was familiar with this feedback in a public address system – too much gain and it screeched! Now positive feedback is the warm and fuzzy type that you get from an ebay customer.
F= G/(1-GH)
Also what is this about destruction of debt/money, is this a legal requirement? is this a bank policy? How is it done? Is it legal?
Cheers
BrightSpark,
1. Difference vs differential equations:
The difference equation (discrete time) approach doesn’t differ from the differential equation (continuous time) approach because these two approaches are (virtually – I don’t want to get into detail) equivalent for all the components of the signal which have frequency lower than the half of the sampling frequency. This is what I wanted to say by mentioning Nyquist-Shannon.
In the frequency domain (assuming that the system is linear) we can use a conformal mapping such as
http://en.wikipedia.org/wiki/Bilinear_transform
to translate continuous-time and discrete-time signal representations and transfer functions.
Please notice that I was using this methodology in digital signal processing so let’s assume that I know what I am talking about.
There are some limitations related to the problem of the uniqueness of the solution of the difference equations but this is not very relevant to our discussion.
NIPA data is sampled either monthly, every quarter or annually. We won’t get better time resolution anyway unless we want to model a stock market crash (transactions happen in a sub-millisecond timescale).
Kalecki / Godley were concerned about processes happening in the time scale of years so his sampling frequency (usually 1 year) was adequate. Personally I would prefer to use the continuous time approach because to me it is easier to think in terms of df(t)/dt rather than (f(t) – f(t-1))/ sampling_period (the Godley’s book is full of indices what makes it particularly hard to read) but we have to keep in mind that any ODE solver internally runs Runge-Kutta so in the end we have a set of difference equations anyway.
2. The issue of money destruction:
When you go to a bank and take a loan for L dollars, they record L on the liabilities side (this is your new deposit) and L on the assets side (this is your new debt). Obviously they will start accruing interests by adding r_L* L to your debt and r_D*L to your deposit every month where r_L is greater than r_D. But imagine that you come to the bank next day after taking a loan and simply change your mind and repay it, L is then subtracted from both the asset and liabilities sides of the bank’s balance sheet (these transactions are recorded in the electronic ledger which is a database system).
The system reverts back to the same state it had had before you took the loan (plus-minus bank fees but this is not the point).
This is exactly an example of the destruction of the credit money. If we want to say that banks can create money by extending credit we also must say that debtors can destroy money by repaying the loans.
Once we acknowledge this obvious fact, we don’t need the non-existing “bank vault” which has been invented to store the money after repaying the debt. Please have a look at the pro-forma balance sheet of Wespac on page 43:
http://www.westpac.com.au/docs/pdf/aw/ic/WBC2010_Annual_Review_and_Sustainability_Report_BKM.pdf
So where is the vault in Westpac? We can see Assets, Liabilities and Equity and by definition Assets = Liabilities + Equity but there is no trace of any “vault”. It is as fake as utility maximization by rational agents in neoclassical economics I would say.
3. Feedback loops:
I agree that there are various feedback loops in the system but in order to model them we must first identify the signal path and the error signal subtraction node. The feedback loop related to the bubble-burst cycle must include asset valuation and credit creation/destruction what is coupled with the productive economy and can affect the level of demand and as a consequence, the level of production/consumption in the economy.
AK
Thanks for your comments.
Agreed you can use difference equations to evaluate an expression which contains integral and differential terms. Agreed this is done in Digital Signal Processing (DSP).
What you cannot do is find the solutions of a differential equation. Having found found the solutions using appropriate methodology the solution equations can then be evaluated using discrete mathematical methodology. Any set of difference equations would yield a different result with the error (ranging from small to infinite) depending on the original equations. Discrete methodology simply cannot correctly model feedback with the error increasing as open loop gain increases.
On the money destruction question we will just have to agree to disagree.
BrightSpark,
Thank you for your response.
I obviously agree with you in regards to the issue of errors in difference equations used to simulate feedback systems with a closed loop. We effectively disregard all the frequency components above f_sampling/2. But in economics/econometrics some of the GDP data (not only in China) is collected with accuracy of several % anyway and so if we analyse the short-term trends we do not make a serious mistake by using difference equations.
http://www.reuters.com/article/2010/12/06/us-china-economy-wikileaks-idUSTRE6B527D20101206
“The U.S. cable reported that Li, who is now a vice premier, focused on just three data points to evaluate Liaoning’s economy: electricity consumption, rail cargo volume and bank lending.
“By looking at these three figures, Li said he can measure with relative accuracy the speed of economic growth. All other figures, especially GDP statistics, are ‘for reference only,’ he said smiling,” the cable added.
Li is widely expected to succeed Wen Jiabao as premier in early 2013, a position that will put him in charge of policy making in the world’s second-biggest economy.”
I also agree that differential equations are much easier to understand than matrices polluted with hundreds of indices. But my point was only that Kalecki-Kaldor-Godley approach is not invalid even if the methods used there can be traced back to the 1921-24 period when Kalecki was studying at Danzig Polytechnic (Gdansk University of Technology).
However the two remaining issues are much more serious. The problem of money destruction is important but is secondary. The most important issue is whether flows depend in principle on flows with some stock dependency (as in Kalecki-Godley) or whether flows depend mainly on stocks due to time constants determining spending / production process lag (what is pure monetarism as in Friedman). So far the models based on the monetary circuit are seem to be monetarist to me.
N.B. production lag for a single company should be modelled as flow_out = flow_in(t-lag) what leads to rather interesting consequences when we want to describe the process in terms of differential equations. This is basically a delay line which may need to be implemented as a Bessel filter, good luck with that…
The analysis presented is oversimplified,but there are some errors and drawbacks which have been discussed and is of great use.
Australia Taxation