Normally, when I write about markets, I try to point at models but there is a lot of guesswork and gut-work in analysis. When times are sort of normal, then models can be a big part of what drives your thinking. But times have not been ‘normal’ for a very long time, and this is part of what drives big policy errors (and big forecasting errors): if you are out of the ‘normal’ range, then to make a forecast or comment usefully on what is going on you need to have a good feel for what the model is actually trying to capture. You need to know where the model goes wrong.

When I was a rates options trader – stop me if I’ve told this story before – I found that I preferred to use a simple Black-Scholes pricing model instead of some fancy recombining-trinomial-tree-with-heteroskedastic-volatility-model. That was because even though Black Scholes doesn’t match up super well with reality, I at least had a good feel for where it fell short. For example, the whole reason we have a volatility smile is because real-world returns have fat tails, but pricing models like Black Scholes are based on the normal distribution. When the smile flattens, it means returns are becoming more like they’re being drawn from a normal distribution; when it steepens it means that the tails are becoming fatter. So that’s easy to understand.

If you understand why an option model works, then it’s easier to think about how to price something esoteric like an option on an inflation swap (which can trade at a negative rate, but actually isn’t a rates product at all but rather is a way of trading a forward price), and not mess it up. But if you just apply and try to calibrate a bad model – especially if it’s really complicated – then you get potentially really bad outcomes. And that is, of course, exactly where we are today.

We haven’t been ‘normal’, I guess, for a couple of decades. Central banks, and in particular the Federal Reserve, have dealt in the markets with a heavier and heavier hand. Nowadays, the Fed not only has expanded its balance sheet by trillions in a very short period of time, but it has expanded the range of markets it is involved in from Treasuries to mortgages to ETFs and now individual corporate bonds. And, since the whole point of this is because the Fed wants to make sure the stock market stays elevated (they are preternaturally terrified at the notion of a wealth effect from a market crash, even though historically the wealth effect has been surprisingly small) I suspect it is only a matter of time before they directly intervene in equity markets.[1] C’est la vie. There is no normal any more.

But at least the ‘normal’ we have had over the last decade was just modestly outside of the prior normal. Things didn’t work right according to the ‘traditional’ way of thinking about things; momentum became ascendant in a way we’ve never seen before and value almost irrelevant. We are now, though, working on a whole different part of the number line. This means that economists will continue to be surprised at almost everything they see, and it means that any model you look at needs to be informed by a good intuition about how the hell it works.

So, for example, let’s consider the money supply. Over the last 13 weeks, M2 is up at a 63% annualized rate. With two weeks left in the quarter, it looks like we will end up with something like a 10.25%-10.50% growth in the money supply for the quarter. The Q2 average money supply, compared to Q1 (important in looking at the MV=PQ equation), is going to be about 13.85% higher. That’s not annualized! Remember, the old record in M2 growth for a year was a bit above 13%, in 1976.

The current NY Fed Nowcast for 2^nd Quarter GDP – keeping in mind that no one has any idea, this is as good a guess as any – is -19.03%. I really like the .03 part. That’s sporty. That would mean q/q growth of -4.75%.

If we want the price deflator to come in around 1.75% (+0.44% q/q), which is where it was for the year ended in Q1, then that means money velocity needs to fall about 16% for the quarter. (1-4.75%)*(1+0.44%)/(1+13.85%)-1 = -15.97%. If money velocity falls less, and that GDP estimate is correct, then inflation comes in higher. If money velocity falls more, then inflation comes in lower. If GDP growth is actually better than -19% annualized, then inflation is lower; if GDP is worse, then inflation is higher. We don’t need to worry much about the M2 numbers themselves, as they’re almost baked in the cake at this point.

The biggest amount that money velocity has ever fallen q/q is about 5%. But clearly, these are different times! We’ve also never seen a 19% decline in growth.

Weirdly, our model has M2 money velocity for Q2 at 1.159, which would be a 15.6% decline in money velocity. Let me stress that that is a total coincidence, and I put almost zero weight on that point estimate. Contributing to that sharp decline, in our model, is the small decline in interest rates from Q1, the increase in the non-M1 part of M2, the small increase in global negative-yielding debt, and (most importantly) a large increase in precautionary demand for cash balances due to economic uncertainty. (This is why it’s hard to get velocity to stay down at this level. The current low levels depend on low interest rates, which will probably persist, but also on dramatic precautionary savings, which are unlikely to). Small changes in money velocity will have big effects on inflation: if our model estimate for velocity was right, we’d see annualized inflation for Q2 at 4.3% or so. Here’s how confident I am in our model: for Q3, it is seeing unchanged velocity (approximately), which with money trends and the GDP Nowcast figures from the NY Fed would imply that y/y inflation would rise to 6.22%, about 17.5% annualized for the quarter. Not going to happen.

Here’s where knowing a bit about the underlying process and assumptions really matters. Velocity is effectively a plug number, in that bureaucrats are good at measuring money and pretty good at measuring GDP and prices, but really bad at measuring velocity directly. So velocity is solved for. And our model (along with every other model, probably) treats the response of money velocity to the input variables as more or less instantaneous. For small changes in these variables – movements in money growth from 4% to 6%, or GDP from 2% to 0% – the assumption about instantaneity is pretty irrelevant. The economy adjusts prices easily to small changes in conditions. But that’s not true at all for big changes. On the available evidence, many prices (if not most) accelerated a bit in Q2, which surprised almost everyone including us. But no matter what the model says, prices are not going to drop 5% in a quarter, or rise 5% in a quarter, for the entire consumption basket. Price changes take time – heck, rents don’t change every month, and it takes time to rotate through the sample. Also, manufacturers don’t tend to make large changes in prices overnight, preferring to drip it in and see consumer response. But here’s the point: the model doesn’t know this. So I suspect we will see money velocity this quarter around 1.14-1.17…not because I believe our model but because I think prices will accelerate by a little bit and I think the real uncertainty surrounds the forecast of GDP. Over time, velocity and inflation will converge with our model, but it will take time.

For what it’s worth, I think that GDP growth will be a little lower than the NY Fed thinks, for a different model reason: the model assumes that changes in various economic data can be mapped to changes in GDP. But that assumes a fairly stable price level…what they’re really mapping this data onto is the nominal price level, and assuming that the price level doesn’t change enough to matter. So I think some of the dollar improvement in durable goods sales, for example, reflects rising prices and not growth, which would be manifested in a slightly lower GDP change and a slightly higher GDP deflator change.

What does this mean and why does it matter?

For one thing…and you already knew this…models are currently trash. They mean almost nothing by themselves. You should ignore it all. I give very little credence to the NY Fed’s forecast. I am pretty sure Q2 GDP growth will have a minus sign, but I couldn’t tell you between -15% and -25% and neither can they. Which is why the -19 POINT OH-THREE is so sporty. But by the same token, you should listen more to the model-builder, and to people who understand what’s going on behind the models, and to people who are taking measurements directly rather than taking them from models. Because this is going back to the art of forecasting, and away from the science. We are over-quanted in this world, and we are over-committed to models, and we are overconfident in models, and we are over-reliant on models. They have a place, just as the autopilot has a place when conditions are placid. When things get rough, you want a real pilot holding the controls.[2]

There used to be a couple of guys in Boston who were auto mechanics and had a radio show. People would call up and describe the noises their cars made, and the guys would ask whether it made the noise only turning left, or both directions, and whether it got worse when it was humid, and other things that sounded crazy to you and me. And then they would diagnose the problem, sight-unseen. Those are the people you want to take your car to. They’re the ones who understand how it really works, and they don’t need to hook your car up to a computer to tell you what the problem is. I took my car to them, and they really were geniuses at it. So look for those people in market space: the ones who can tell by the sound of the squeal what is really going on under the hood. They won’t always be right, but they will have the best guesses…especially when something unusual happens.

[1] Ironically, I think that something else they are considering would have a much bigger effect on equity markets than if they directly bought equities, but I don’t want to talk about that in this space because it also has big implications for inflation-related markets and would create some really delicious relative value trades that I don’t want to discuss here.

[2] Although I didn’t think I’d remark on this in today’s comment: this is also why the Trump Administration’s move today to loosen the Volcker Rule to let banks take more risks with their capital is very timely. There is a lot of bumpy flight ahead of us and we should want seasoned traders making the markets with actual capital behind them, not robots looking to scalp an eighth.