The ability to predict inflation

[mooncliff]

This article starts out in mathematics and physics, and then passes through many other sciences, but winds up saying that it may not be possible to reliably model inflation more than a month or two in advance.


To find out, Bouchaud and his colleagues looked at how well US inflation rates could be explained by a wide range of economic indicators, such as industrial production, retail sales, consumer and producer confidence, interest rates and oil prices.

Using figures from 1983 to 2005, they first calculated all the possible correlations among the data. They found what seem to be significant results - apparent patterns showing how changes in economic indicators at one moment lead to changes in inflation the next. To the unwary observer, this makes it look as if inflation can be predicted with confidence.
...
In recent years, some economists have begun to express doubts over predictions made from huge volumes of data, but they are in the minority. Most embrace the idea that more measurements mean better predictive abilities. That might be an illusion, and random matrix theory could be the tool to separate what is real and what is not.


It is fascinating how in the era of huge data sets, biology becoming a computational science, etc., patterns are beginning to emerge across the data sets. I think this is potentially really useful because it suggests that patterns in data sets in the experimental sciences can be used as guides about the reliability of models in sciences in which it is difficult to do experiments.

http://www.newscientist.com/article/...html?full=true

[c1ue]

Re: The ability to predict inflation

It is a good article - until it starts talking about inflation.

Inflation is different than fundamental laws of physics. Inflation can be deliberately increased or decreased by monetary policies.

To use methods for determining the boundaries of natural laws to predict inflation is to believe that there is no free will. But we do know that Zimbabwe was a choice, much as the present course of the United States is a choice.

[jtabeb]

Re: The ability to predict inflation

It is a good article - until it starts talking about inflation.

Inflation is different than fundamental laws of physics. Inflation can be deliberately increased or decreased by monetary policies.

To use methods for determining the boundaries of natural laws to predict inflation is to believe that there is no free will. But we do know that Zimbabwe was a choice, much as the present course of the United States is a choice.

Correct.

[bart]

Re: The ability to predict inflation

My prediction work is far from perfect, but it's workable:

[charliebrown]

Re: The ability to predict inflation

Bart, how does your CPI-U, prediction v1.2 work? Be general.
I assume if the default issues can be swept under the rug for awhile longer this could be bearish for PM's. This would suit me just fine.

[ThePythonicCow]

Re: The ability to predict inflation

To use methods for determining the boundaries of natural laws to predict inflation is to believe that there is no free will.

Pashaw.

The basic difference between humans and the lessor living creatures we know is that humans tend to confuse the divine and the ordinary. We do indeed have more powerful minds, better able to notice the divine, but we also have a more complex psycho-dynamic structure, capable of more elaborate confusions. This is closely related to the reason humans can feel empathy and shame, whereas lessor creatures at best feel only sympathy and guilt (reptiles not even those.)

The human mind can sense some of the more subtle order of the universe, and the human psyche can "give spirit" to that order, either imagining some supreme being(s) (super-humans able to create and order the universe) or some super power of humans (free will for example) as if those powers lifted us above the ordinary laws of nature, and random matrices that can model them ;).

Whether or not some particular set of measured values can be usefully analyzed using random matrices is a quite different and rather more trivial question. All that's going on here is that there are certain fairly interesting statistical properties of collections of statistics when in the form of random matrices. That is, such matrices are a useful model. A small amount of work to shape your data into that form yields a surprisingly large amount of additional knowledge about that data.

Whether that is useful or not obviously depends on whether your initial data "span" (form a sufficiently complete set of eignenvectors of) the space being analyzed. The economics statistics we have are woefully inadequate in this regard, it seems. Bouchaud's result applies Marčenko-Pastur's Law, using the very sorts of properties of large random matrices that make them so useful to also detect when they are not so useful. These results do indeed confirm the inadequacy of our economic statistics. In such cases, piling on more of the same statistics is just piling "it" higher and deeper.

[bart]

Re: The ability to predict inflation

Bart, how does your CPI-U, prediction v1.2 work? Be general.
I assume if the default issues can be swept under the rug for awhile longer this could be bearish for PM's. This would suit me just fine.

We started out with the basic definition of inflation. We then applied it by locating all sorts of different goods and money data from many sources, including but not limited to all the ones mentioned in our definition of money measures. As a rough guess, about 80% of the data comes from the FRED database at the Federal Reserve site here. Then we plugged all that data into a set of huge Excel© computer spreadsheets.

Next, by a very long process of trial and error, plus adding different time lags and weights to the various numbers, we came up with the basic predictions. By weights, we mean that items like Bank Credit get a much higher weight than for example Currency since the amount of credit available is much more important to money growth than printed currency in circulation. By time lags, we mean that different types of money take differing amounts of time to get into the economy. For example again, tax refunds or Fed Open Market Operations get into the economy faster than changes in a money measure called M3 (M3 is defined in money measures).

Lastly, in order to modify the basic prediction model to match a particular market like the stock market, we did more research and located factors that specifically affect it and then added them into the prediction matrix. One example that's part of the set of stock market adjustments is program trading percentages and dollars (data available here).

The whole process started from being curious to see if the basic definition of inflation and deflation could be directly translated into predictions, and we were quite surprised at the results too. Do note that like almost anything, not only are they not 100% accurate but also eventually the formulas will cease to work well at which time we'll go back to the drawing board. Also note that there are at least two significant missing elements in all the predictions - any objective measure of human fear and any allowance for unexpected events like terrorism or calamities, etc.

(Note for folk with a large economic background: there are many adjustments and factors for more exotic methods of money creation represented by GSEs, ABSs and MBSs, hedge fund activity, fractional reserve banking, etc. There are also adjustments for GDP hedonic factors on the goods side.)
(last update mid 2005)


(from my FAQ page)

[charliebrown]

Re: The ability to predict inflation

thanks. very helpful. I'm checking out your web site today.

[c1ue]

Re: The ability to predict inflation

We started out with the basic definition of inflation. We then applied it by locating all sorts of different goods and money data from many sources, including but not limited to all the ones mentioned in our definition of money measures. As a rough guess, about 80% of the data comes from the FRED database at the Federal Reserve site here. Then we plugged all that data into a set of huge Excel© computer spreadsheets.

Next, by a very long process of trial and error, plus adding different time lags and weights to the various numbers, we came up with the basic predictions. By weights, we mean that items like Bank Credit get a much higher weight than for example Currency since the amount of credit available is much more important to money growth than printed currency in circulation. By time lags, we mean that different types of money take differing amounts of time to get into the economy. For example again, tax refunds or Fed Open Market Operations get into the economy faster than changes in a money measure called M3 (M3 is defined in money measures).

Lastly, in order to modify the basic prediction model to match a particular market like the stock market, we did more research and located factors that specifically affect it and then added them into the prediction matrix. One example that's part of the set of stock market adjustments is program trading percentages and dollars (data available here).

I have the utmost respect for your work, bart, but I would note that the definition of prediction in this case is very specific: that the CPI-U number is a function of various amounts of Fed and other monetary activity.

The latter can be said to be an expression of monetary policy. I note that prediction of monetary policy is not covered by the graphs.

Perhaps you could comment on the lag time between changes in the input amounts and a change in the inflation number(s).

This would clarify the difference between reading tea leaves vs. looking at a satellite view to predict the weather...

[KGW]

Re: The ability to predict inflation

Predict inflation? Easy. Fiat currency=Inflation. Kind of like when it rains the ground gets wet. . .The rentier class has gotten its hoard by building in inflation. FIRE is simply the "fruition" of their scheme. Have you looked at the price of most things, relative to the income levels of the masses? In order to buy most of these things, people are encouraged to take on debt, as if it were the natural order. Therefore, interest payments to the rentier class have increased, like blowing up a ballon. ;)

[bart]

Re: The ability to predict inflation

I have the utmost respect for your work, bart, but I would note that the definition of prediction in this case is very specific: that the CPI-U number is a function of various amounts of Fed and other monetary activity.

Indeed - per the basic over simplified inflation definition of more money than goods.

The latter can be said to be an expression of monetary policy. I note that prediction of monetary policy is not covered by the graphs.

It's implicitly covered, since monetary and fiscal policy is reflected directly in money supply figures.

Perhaps you could comment on the lag time between changes in the input amounts and a change in the inflation number(s).

This would clarify the difference between reading tea leaves vs. looking at a satellite view to predict the weather...

Money supply factors and lags

Item	Lag, operable time frame	Notes
M1	4-10 months, averages 6-9
M2	9-20 months, averages 12-18
M3	9-20 months, averages 12-18
Monetary base	6-12 months, averages 8-16	High importance, controlled directly
Fiscal (government) stimulus	9-24 months, averages 12-18
Bank credit	Two lags - one is 3-6 months, and the other is similar to M3	High importance
Fed & comm'l repos	Two lags - one is virtually immediate and is more important, and there is also a secondary lag of about 10 months	High importance to stock market - the best single stat proxy for the Fed
Securities Lending	Two lags - one is virtually immediate, and the other varies between 1-4 months	High importance to bond interest rates

The changing velocity of money is the primary reason for the wide range in lags from the time money is created until it is reflected in inflation. With high inflation, money moves faster and there appears to be more in the system. Note that these numbers are only rough guidelines.

Also note that the lags above only apply to the U.S. Federal Reserve. Other countries central banks may have different lags since money measure definitions differ.




"Significant changes in the growth rate of money supply, even small ones, impact the financial markets first. Then, they impact changes in the real economy, usually in six to nine months, but in a range of three to 18 months. Usually in about two years in the US, they correlate with changes in the rate of inflation or deflation.
The leads are long and variable, though the more inflation a society has experienced, history shows, the shorter the time lead will be between a change in money supply growth and the subsequent change in inflation."
-- Milton Friedman, economist




(from my money & lags page)

[cmalbatros]

Re: The ability to predict inflation

I wonder if the Hydraulic Econometric machine is still being used at the London School of Economics ?