If conditions are right, a small rabbit jumping across the snow could set off a huge avalanche. Conditions in Japan are such that some small event could trigger hyperinflation at any time. Maybe China sinking a Japanese boat or a pension fund getting out of JGBs. We don't know what the equivalent of the rabbit will be for Japanese hyperinflation, but we can tell Japan has the setup for this type of chain reaction.
Again and again there are blog posts where people say the fact that Japan or the USA has not had hyperinflation so far proves that the theories saying there is a risk of hyperinflation are wrong. This is just not true. That an avalanche has not happened yet is in no way proof that there is no risk of avalanche. Such logic is flawed. It ignores the fast moving chain reaction that is fundamental to the nature of the problem. The same is true for hyperinflation. That it has not happened yet does not in any way prove that there is not a high risk of hyperinflation. Many countries went from normal inflation to hyperinflation rather quickly.
I have yet to have any really strong logical attack on my hyperinflation explained or hyperinflation faq posts. More than 100 times I have tried to engage other bloggers posting about hyperinflation.
The lack of any serious intellectual challenge to my "positive feedback theory of hyperinflation" is more of a confirmation of my theory than the lack of hyperinflation in Japan or the USA is any sort of refutation.
Cnningham's Law says, "the best way to get the right answer on the Internet is not to ask a question, it's to post the wrong answer.". If I was wrong, I really think someone would have been able to point this out by now.
The xkcd Duty Calls sort of sums this up:
I think it is safe to say there is nothing obviously wrong with my theory of hyperinflation.
Dear reader, if you think there is anyone that could refute my theory of hyperinflation please send them a link to this post.
Vincent, you are hypothesizing that hyperinflations are random events, but that we can estimate the probability for them happening. The following would be very impressive for your theory or at least your personal ability to predict things (but most impressive would be a theory, will formulas to calculate probabilities). So why no do the following: There are approximately 200 nations on Earth, most of which (all?) use money, and thus can experience hyperinflation. Why not run your theory on as many of them as you can and construct a table with the probability for hyperinflation over the next year, over the next two years, three, etc, out to a decade or so. Then we'd all have a numerical means of grading your theory. Again, if your theory were expressed as a formula so that us people at home could duplicate your probability calculations (and maybe even help you with those calculations), then that would be ideal.
ReplyDeleteThe more countries and years you include in your forecast the better we can assess the predictive power of you hypotheses. Will all that prove anything? No... but it can say whether or not your model is more or less likely true.
Also, it would be fun to compare with some sort of random means of assigning probabilities to see how well you stack up against the "darts" as the WSJ used to do on a regular basis with stock pickers (the darts usually did very well by comparison, if I recall correctly). You could have a couple of different "darts" algorithms actually... I can think of a couple off hand.
That would be very interesting, and it would be a means of putting your ideas to the test against nature. Let nature be the judge! :D
I can point you to some specific predictions of inflation in different countries over the next 1 to 6 years. No hyperinflation was predicted, so maybe you're not interested.
As a simple example, say you calculated that the US has a probability of hyperinflation of P over the next year, and that this is independent from year to year, but always P. Then the probability that the US would have hyperinflation sometime over the next decade would be 1-(1-P)^10. If P = 0.3, then this is 0.97 probability the US will have hyperinflation sometime over the next 10 years, 0.995 over the next 15 years, and 0.9992 over the next 20 years. I realize they may not be independent, but this is just a simple example. With numbers like that we could sort out how well your theory matches reality with some confidence.
DeleteIf you don't like my idea, why not look up how plate tectonic theorists statistically predict the likelihood of earthquakes in different regions, or something similar from natural sciences. I'd be shocked if such predictions don't exist (the probability of an earthquake in New York has GOT to be lower than for California). Perhaps you can emulate what they do. Let's bring science into this!
DeleteMy problem with your theory as it stands is circularity. The MM guys are guilty of this because their argument goes something like this:
ReplyDeleteThe QTM would function exactly like the formulas predict except that X is also required. How do you know if there's enough X? There's no way to know except if the QTM is working. (BTW X = "expectations" usually).
You cannot refute this argument. It's logically consistent and consistent will all observed data. Yet it explains nothing, and thus Occam's razor dictates that it should be "cut" out of the picture. Now if there were an independent means of measuring whether or not X was sufficient outside of observing if the QTM is working, then that would be a big improvement: then we could check the X hypothesis against nature to see if it's false.
I realize you've made one Japan prediction, but that gives us little to work with. Even if your prediction comes true, that's not enough data points to really judge your theory. That's why it would be much more impressive if you constructed a table with 100 countries, and the probability for each. Again, if you just devise the formulas, then your loyal readership can help you collect the necessary numbers and calculate them and check each other's calculations even.
So what I'm saying is you can't shoot down the MM argument either. One could easily rewrite your post as if Scott Sumner wrote it about MM: I'm sure they've had people try multiple hundreds, perhaps 1000s or 10s of thousands of times to poke logical holes in their arguments. Why is their theory so "successful" and impervious to "logical arguments?" Because it's based on circular reasoning! (Not to pick on Scott or MM really, because I think this goes on a lot in macro: nobody is very willing to let nature sort out if their theories are true or not: nobody writes something like "If A happens then I'm wrong"... and who knows, maybe someday we can directly measure "expectations" with brain implants and really put their theory to the test).
DeleteThe test is not logic, it's nature: make predictions, even statistical ones, and let nature decide. If they are statistical, then you need to either wait a long time or make a lot of them in parallel to get the results back.
I can easily shoot down the MM main idea of NGDP targetting. It fails when the government needs money and gets the bank to keep buying bonds even when their original plan would say to stop. This is the same way all the other central bank plans fail.
DeleteConditions for a positive feedback loop is not the same thing as circular logic. There is no circular logic in avalanche theory.
DeleteImagine someone found that stock market crashes usually happened after margin debt was over some percentage. Further assume it was true that margin calls were a big part of the crash. The more the market went down, the more margin calls. But the more margin calls then the more people had to sell. So there is a positive feedback loop leading to destruction. Now if margin levels were very high we could then say that the risk of this positive feedback loop of stock market crash is high.
DeleteIt is a similar thing with debt and deficit levels. When they are high there is risk of a positive feedback loop that is hyperinflation. This is not circular logic.
But if you use the Avalanche argument to claim that "we won't know if it happens until it happens: when it does happen I can point to what happened and say 'Ah, the risk was high!' and if it doesn't happen, I can say 'Ah, the risk is high, but it just hasn't happened yet'" then I claim that's a useless description. I can just as well claim "The risk is low, but sometimes there is a high risk, but you don't know it's high until it happens" then when it doesn't happen I can claim "Ah the risk is low, thus it doesn't happen" and when it does happen I can claim "Ah, the risk was clearly there in that case" .... but again, this is unhelpful.
DeleteYou say:
"Now if margin levels were very high we could then say that the risk of this positive feedback loop of stock market crash is high."
Let's put this to the test with some numbers to quantify the risk. Numbers that will help us sort out (as events happen in the future) how well these numbers described the risk you say is present now.
Also, I wasn't using NGDP targeting as the main MM idea. I'm not convinced you've shot down NGDP targeting anyway. But I've read your description a couple of times, and I don't see what you mean. But back to what I was talking about regarding MMs: they essentially say that monetarism works just like QTM models suggest, except that the mystery ingredient X (expectations) needs to be present too, but the only way to know if it's present is if the QTM works. That's that part that can't be logically assaulted: it covers all the bases by assuming the conclusion. But that's why it's useless as well... at least until there's an independent way to measure X and put it to the test.
That's like hypothesizing that coin flips *always* come up heads provided sufficient X is present. When tails come up I simply say X was missing. And the only way to know if X is present is to flip a coin. Again, the X hypothesis is perfectly logical and useless.
If you understand the chain reaction for an avalanche, forest fire, earthquake, or hyperinflation then you can understand when conditions are right for the chain reaction. It is not just a useless description. Tom, you are wrong and not giving any good argument.
DeleteGive a good set of numbers for the risks, then we can put your theory to the test.
DeleteI made a follow up post to this to address Tom's charge of circular logic.
Deletehttp://howfiatdies.blogspot.com/2014/08/positive-feedback-theory-of.html
Vincent, I'll say that your positive feedback argument sounds plausible, but I'll also say it sounds like you're overestimating the risk (it's hard to tell, since you don't give a specific quantity of risk).
ReplyDeleteSo two people could share your theory and have very different conclusions based on different quantizations of risk. So again, unless you put numbers for the risks, your theory doesn't even help distinguish expected outcomes: depending on how risk is quantized, all outcomes can be consistent.
I could make a similar sharia law prediction: I can say it comes like an avalanche. But the trick is to assign numbers. I will gladly assign numbers to 100 states and provinces of various countries for the probability that they go sharia over each year for the next 10 years. Since I don't have any passionate ideas about sharia law and just came up with that example just now, I won't be providing a numerical model... I'll just assign the numbers from my "gut feeling." But I'll be glad to do it. I'm sure my numbers will differ wildly from other people... I'd put the probability for N. Carolina at 1e-10 for example, but somebody out there thinks it's more like 50% I'm sure. In 10 years time, my global expected sharia law map will look very different than theirs, even though we may share the underlying "avalanche" theory of going sharia.
Actually, I'm prepared to take my sharia law avalanche theory to the next level: I'll say the probability each year of a state or province going sharia (if it's not already sharia) is equal to the sum total of the population which identifies themselves as Muslim divided by the total population.
DeleteWe should have enough information in a year's time to decide if my sharia avalanche hypothesis is likely false. In ten years we can make a much better determination.
Don't forget that the population should first be disarmed. The USA spent years searching house to house in Iraq taking away peoples guns. Then a group like ISIS can get a big stash of weapons (left by USA) and roll through and with lots of guns and local support give you your avalanche. Even if Arizona was 50% Muslim it would not avalanche because there are so many guns.
DeleteI don't have the historical data that would let me put probabilities even on 3rd world countries. But modern western world reserve currency hyperinflation has no historical data to compare too. In the past world reserve currencies were tied to gold or silver or were gold or silver. Also, this fundamentally is a human thing and humans can change a lot of stuff in unpredictable ways. So it is just not easy to put a number on, at least for me so far. But even if I could not put a number on real avalanche probabilities or forest fire probabilities it would help to understand the theory of positive feedback loop.
DeleteVincent, I'm sympathetic the to difficulty of assigning numbers, but I've seen it done (in other macro contexts), so I know it can be done. It's really the only way to make any progress in my view. Make testable predictions: the more the better.
Delete"Also, this fundamentally is a human thing and humans can change a lot of stuff in unpredictable ways."
True, humans are involved, but that's no excuse for why the scientific method can't be applied. Just how much stock we should put in the special "human nature" of macro though is up for debate: [1] [2] [3] [4]
What I like about Jason's approach is he starts with core hypotheses, adds some auxiliary hypotheses, creates mathematical models from the hypotheses, demonstrates how his mathematical models fit past data, demonstrates how he can "forecast" out of sample past data from earlier past data using his models, and then he goes ahead and puts it all on the line making specific, testable numeric predictions of the future based on those models, and then states that's he's not much interested in adding epicycles to his hypotheses should the tests fail: he'd rather declare his hypotheses and models false, write a postmortem on his blog, and move on. I don't know of any other econ bloggers out there willing to go out of their way to expose their hypotheses and models to falsification by nature like that, do you? Mostly they seem to go out of their way to protect their hypotheses and models from nature's judgement. Most are probably rhetoricians trying to convince people of questionable and sometimes ill-defined opinions (perhaps ultimately based on "gut feelings" or intuition?) rather than scientists trying to find the truth. If there's one thing that science should have taught us over the past couple of hundred years it's that human intuition and "gut feelings" are frequently lousy judges of what's actually true. Which is why I have a problem with statements like this:
"It is a similar thing with debt and deficit levels. When they are high there is risk of a positive feedback loop that is hyperinflation."
At least without the rest of the scientific rigor to go along with it: That sounds like a gut feeling to me, not an evident fact. That's fine, but that carries no weight in deciding if it's true or not. Let's figure out a way to let nature decide if this hypothesis is true or not.
BTW, you brought up the debt thing to Jason on his blog and he responded. I don't know if you saw that or got back to him. Basically he said he saw no correlation in the past data to justify bringing debt into his hyperinflation model.
If there are not a sufficient number of testable predictions made, how are we to decide if it's you that's the unreasonable fear mongering Chicken Little, or it's your ideological opponents who are being willfully blind Pollyannas? All sides need to go on record with their predictions. Then we'll let nature judge. Science is fun for those of us not shouldering the burden of proof which is the responsibility of all those making hypotheses to shoulder. :D
DeleteWith regular inflation there might not be a correlation to debt and deficit over a short enough time period. However, they are always very high ahead of hyperinflation.
DeleteBernholz studied a bunch of hyperinflations and the data shows that high debt and deficit are the best predictors for hyperinflation. Link off to the right. It is not just a gut feeling.
Tom, if I had it down to where I knew the numerical probability of each country getting hyperinflation over the next year, I would not be trying to get into debates on my blog. I would be selling everything and gambling in the markets.
DeleteMy hypothesis: For most people
Deletegambling in the markets confidence > making a testable prediction confidence
Tom, I am gambling in the markets that the Yen will go down, just not at the "sell everything I own" confidence level.
DeleteVincent, why is your approach superior to this model?:
ReplyDeletehttp://www.cemla.org/PDF/investigacion/inv-2013-04-08.pdf
My simulation does not require complicated math.
DeleteMy simulation uses real things, debt, deficit, money supply, velocity of money, etc that are concrete and easily understandable.
My simulation shows the feedback loops, how not rolling over bonds means more central bank buying of bonds, etc that really happen in hyperinflation. Shows the how and why of hyperinflation.
Also I believe that you have had significant push-back on your avalanche theory: Mark Sadowski made a compelling argument that 1st world countries always have multiple years of high inflation prior to hyperinflation. What's high? I'd have to re-read Mark for clues, but off-hand I think high single digit and low double digit qualify. He also pointed out that high debt levels and deficits in 1st world countries are not correlated with hyperinflation or high inflation: in fact just the opposite is true: low debt levels and low deficits are correlated with high inflation (take for example, the highest inflation rate post-WWII in the US: about 1980 or so: that was one example Mark gave).
ReplyDeleteMark does not count 26% as hyperinflation. If you only count 50% per month then it might take a year or two to get there. But the move from 2% to 26% is the interesting part.
DeleteHe does not have any problem with the basic idea that it is a positive feedback loop.