Archive
Inflation Guy’s CPI Summary (November 2025)
What better way to end this crazy year than with an economic data point that we don’t know how to really interpret? Happy New Year!
Recall that, thanks to the government shutdown, the BLS released September CPI (by recalling workers to calculate the number based on data already collected) but didn’t do any of the normal price-collection procedures for the prices that are normally collected by hand. That’s far less than 100% of the index, but it’s a lot and so the October CPI was not released at all. Which brings us to today, and the November CPI – where the data was mostly collected somewhat normally. However, the calculation procedures had to be adjusted in ways we don’t really know about. You’d think that the way you do this is that you figure out the value that equates to the price level you just measured, and just say ‘hey, that’s a two-month change’ but it isn’t quite that easy. And some very smart people think this could bias the CPI lower for a few months. Whatever they end up doing, the lack of an October number is still going to mess up all the feeds (e.g. from Bloomberg) and all of the scripts and spreadsheets based on those feeds.
The BLS said in a FAQ yesterday that “November 2025 indexes were calculated by comparing November 2025 prices with October 2025 prices…BLS could not collect October 2025 reference period survey data, so survey data were carried forward to October 2025 from September 2025 in accordance with normal procedures.” In other words, November will basically be a 2-month change. (Or so we thought: see below).
Looking back to the last real data we got, in September: recall CPI was weaker than expected, but a big part of that was because of what looked like a one-off in OER. But the breadth of the basket that was accelerating was increasing, which was not a good sign. Normally the OER question would have been answered last month but…oh well.
Coming into the month…we at least have market data!
There was a big drop in short inflation swaps and breakevens this month. A lot of that is due to the steady drop in gasoline prices (see chart below), but some of it may be because sharp-penciled people anticipated that the BLS adjustment for October’s missed data is going to bias the number lower.
And boy, did it. This number is absolute garbage.
There are going to be two eras going forward: pre-shutdown inflation data and post-shutdown inflation data. Much like when there are large one-offs in the data, as in Japan years ago when there was an increase in the national sales tax rate, the year-over-year data for the next year are going to look artificially low. The BLS never adjusts the NSA data ex-post. If it’s wrong, it stays wrong. We can really hope that this doesn’t affect seasonal adjustments when the BLS calculates the new factors for next year, because that would mean next October’s CPI is going to be massively biased upwards.
Because what it looks like is that for many series the BLS didn’t calculate a two-month change based on the current price level – it looks like, especially for housing, they assumed October’s change was zero so that the two-month change reported for this month was actually a one-month change spread over two months. For example, even with the low Owners’ Equivalent Rent print in September, the y/y figure was 3.76%, so about 0.31% per month. The BLS tells us that the two-month change in OER was +0.27%. That looks more than a little suspicious to me.
Largely from that effect, core services inflation dropped from 3.5% y/y to 3.0% y/y in just two months. Riiiiight.
If in fact these two-month changes are all (or mostly) one-month changes, then the data makes a lot more sense. Either way, it’s hard to believe that the y/y change in Health Insurance dropped from 4.2% y/y to 0.57% y/y, thanks to a -2.86% decline in November from September. Yes, the Health Insurance category does not directly measure the cost of health insurance policies, and October is often when the new estimation from the BLS goes into effect, but a monthly -1.43% pre month decline for the next 12 months in Health Insurance is implausible.
Ergo, I’m not going to show most of my usual charts. This is garbage all the way down. Now, in my database instead of having a blank for October as the BLS does (for many but not all series. Seriously this is going to completely mess up any spreadsheet based on pulling data from Bloomberg), I am going to assume the price level adjusted smoothly over those two months – that is, I interpolated between September and November. That’s naïve, but it’s necessary to assume something and that’s better than assuming no change for October!
I have no idea what this will do to Median. If the Cleveland Fed follows the BLS lead, they’ll report a blank for October and a Median of something like 0.24% for the two-month period (that’s what I calculate), but it’s also garbage because garbage-in, garbage-out.
Really, this is a low point for inflation people and a low point honestly for Inflation Guy. I expected more from the BLS. I spend a lot of time defending these guys (heck, I just wrote a column on “Why Hedonic Adjustment in the CPI Shouldn’t Tick You Off”) because the staff involved in calculating the CPI are solid non-partisan professionals (aka pointy-head types) who really are trying to get as close to the ‘right’ answer as actual data allows. I can’t say that’s true in this case. Now, maybe when we get more data we will discover that the economy has abruptly shifted into something like price stability on the way to outright deflation, and it just happened to have a major inflection in October when no one was looking. But to me, it just looks like bad data.
Policymakers still gotta make policy, even if garbage data is all they have. But the correct response to not knowing what’s happening is not to assume you know what’s really happening and act accordingly – the right approach to extremely wide error bars is to do nothing. The correct approach for the Fed is to do nothing until they have another 3-6 months of data and can start getting some confidence about current trends again. That’s not the world we live in. In this world, the Fed will recognize that the inflation data is squirrelly so their behavioral response will be to ignore it and in the policy context that means that they’ll make policy for a while here based solely on the labor market. Get ready for much more market volatility around the Payrolls report again! To me, that looks like it’s likely to be an ease in two of the next three meetings, before the FOMC needs to recognize that the new inflation data is still showing 3-4% inflation. It’s possible that the Committee could take a pause while they wait for the incoming Fed Chair in May. But the inflation data will not be an impediment to an ease, and will no longer be a strong argument for holding the line if growth data looks weak.
I may be being overgenerous here. It’s also possible this will reinforce the FOMC members’ priors since many of them were utterly convinced that inflation was going to drop significantly due to housing. This, in the presence of bad data, would be a pure error. But the result is the same: an easier Fed than is healthy for the monetary system right now.
There are lots of reasons to think that yields further out the curve will stay stable or rise. But yields at the short end should probably reflect easier money going forward.
Sorry I couldn’t be more help. Here’s looking forward to 2026!
Why Hedonic Adjustment in the CPI Shouldn’t Tick You Off
I’ve worked in the inflation field for about a quarter-century (depending on how you want to count it), and I can tell you that if you really want to start a food fight at an investment conference, mention the term ‘hedonic adjustment’ as it relates to the Consumer Price Index. Thanks to substantial counter-programming by people who want you to prefer their narrative on inflation and their inflation index, people who tend to hold to the “the government is making it up” narrative about inflation like to quote hedonic adjustment as one element of proof.
The first problem with this is that people seem to think that CPI is supposed to measure how their actual cash costs change every year. It isn’t. If you look at the price of anything, it represents a trade offered by the supplier of value for value: if you give me X dollars, I will give you the widget that paints your house in 6 hours. If you don’t think that widget is worth X dollars, then you don’t buy the widget.
But widgets change. If the same vendor offers the same widget, but thanks to improvements now will paint your house in 3 hours, and now costs Y dollars, you the buyer have the same evaluation to make except now it’s the value of a 3-hour paint job versus Y dollars instead of 6-hours versus X dollars. If you want to see how the trade changed, then you can’t just compare Y versus X. You have to compare the other side of the trade also. Or, to put it another way, the difference in price (Y-X) isn’t just due to the fact that the dollar is worth less now than it was, so that even the old version of the paint-widget would cost X’, but also because it’s a better widget. You the consumer see the price going up from X to Y, but that consists of inflation X’-X, plus quality improvement of Y-X’.
There are no two ways of looking at that. If you want to measure the change in cash outlays, just count your cash outlays. But if you’re trying to measure the change in the cost of living, then you need to try to hold the standard of living constant between measurements.
So any inflation measurement needs to account for the fact that widgets change, or it will perpetually exaggerate inflation.
Most of those adjustments are pretty straightforward. If your candy bar got 20% smaller, it’s easy to account for the additional inflation that implies. In fact most of these quality adjustments are called “quality adjustments.” It becomes a ”hedonic” adjustment when the widget has a lot of different elements that give it value. Think of a car, where having better fuel efficiency is valuable but so is an improvement in the dashboard entertainment system. When the price of the car changes, it’s much harder to figure out how much of that due to inflation (paying more to get the same stuff, X-X’ in the example above) and how much is due to the change in the components of the vehicle. Enter the econometrician, who applies fancy mathematics that you may be unsurprised to learn is called a ”hedonic regression.”
Now, just about 100% of the CPI basket is subject to quality adjustment when necessary. As I said, quality adjustment is necessary. But only a small fraction of the basket is adjusted using hedonic regression.
But it’s not even as bad as that. You hear a lot of grumbling about how “hedonic adjustment says the price of a computer is falling even though it’s staying the same or going up, so obviously inflation is really higher than the government says it is.” But you almost never hear anyone complain about hedonic adjustment to shelter. The BLS, you see, adjusts for the fact that the housing stock gets older, so that if you pay the same rent from year 1 to year 2 it actually works out to be inflation because you’re getting a slightly older apartment. The real kicker? The upward hedonic adjustment to shelter inflation comes very close to balancing the downward hedonic adjustment to computers and microwaves and things. In other words, if you outlawed hedonic adjustment it wouldn’t really change the CPI hardly at all. A 2006 paper by Johnson, Reed, and Stewart found that the “net effect of hedonics from 1999 onward…is estimated to be less than 1-hundredth of 1 percent per year, specifically +0.005 percent.”[1]
So honestly, the bottom line is that people yell about hedonic adjustment for the same reason they yell at referees. They have to yell at something when they don’t like the outcome!
Is hedonic adjustment “right?” That is, does it correctly determine how much of a price change is due to inflation and how much is due to quality changes? I can say with great certainty that it is not exactly right. It’s an estimate. Virtually every financial model is an estimate. The Black-Scholes option pricing model isn’t right either – in fact, we know that the Black-Scholes model isn’t just wrong, but it’s wrong in some very systematic ways. And yet, people continue to use Black-Scholes, because we understand the ways in which it’s not right and can adjust for it.[2]
Hedonic adjustment is also not “right.” But it’s a fair approach, and if you want to adjust the CPI by removing the downward hedonic adjustments while keeping the upward hedonic adjustments (to shelter) then you can make that adjustment mentally by just adding about +0.10% per annum to the CPI. Either way…it shouldn’t tick you off.
[1] Johnson, D.S., S.B. Reed, and K.J. Stewart. 2006. “Price Measurement in the United States: a Decade After the Boskin Report.” Monthly Labor Review (May): 10–19.
[2] One big way is that since actual market movements aren’t distributed normally, and the Black-Scholes model assumes they are, the price of options that are far out-of-the-money are systematically low. Or they would be, if we didn’t adjust for this known problem by applying a volatility smile to price out-of-the-money options.
Modeling Shortfall Risk versus Inflation – What a Good Hedge Looks Like
When people ask me about hedging inflation, they aren’t always asking what they think they’re asking. There are two approaches to addressing inflation in your portfolio so that the portfolio grows in real terms. One of the approaches is to try to simply outrun inflation: “If inflation averages 3%, and I have an investment that averages 5%, I’ve succeeded.” This mode of thinking derives, I think, from the fact that all of our education has been in nominal space and in most financial modeling problems inflation is just assumed rather than modeled as a random variable. It turns out to be a lot harder than it sounds to find an asset class or collection of asset classes that dependably beat inflation over moderate (10+ year) periods, because there is significant (inverse) correlation between inflation and the performance of many asset classes. Most obvious here are stocks and bonds, so if you build a 60-40 portfolio that “should” return 5% over the long term and figure that will beat inflation, you’ll be right…as long as inflation stays low. If inflation goes up, you won’t only lose purchasing power but you’ll lose actual nominal value, since equities and bonds both tend to decline when inflation goes up. Let’s put that aside for a second but I will come back to it.
The other approach to addressing inflation is to try to hedge inflation: exceed inflation by a little bit, but all the time, so that your returns go up when inflation goes up and your returns go down when inflation goes down, but you always are experiencing some positive real return.
The difference between the first approach and the second approach can be summarized by thinking about shortfall risk. As an investor, you care about the upside (in real terms) but most of us are risk-averse meaning that we care more about the downside. Ask most people whether they’d risk a 25% loss in their portfolio purchasing power to have a similar risk of gaining 25%, and they will experience a strong preference to avoid that coin flip. Risk aversion isn’t linear, so investors treat small gains and losses differently from large gains and losses, and of course it matters whether you’re barely covering your goals or easily exceeding them so that you’re ‘playing with house money.’ Many things, in other words, affect risk preferences. But the bottom line is that if you are trying to ‘hedge’ inflation, you care about your shortfall risk over some horizon. What is the probability that you underperform inflation – that is, lose value in real terms – by some given amount between now and a stated horizon?
Now we are going to get a little mathy, but for those who aren’t so mathy I will try to explain in English as well.
If you want to evaluate the probability of asset B underperforming asset A by some given amount over some period, of course you need an estimate of the expected returns of A and B, or how they’re expected to drift relative to one another. That determines your jumping off point. Let’s suppose that A and B have the same expected return. The next thing that determines the frequency and severity of a shortfall of B versus A is the volatility of the spread between them, which is driven by (a) how correlated A and B are, and (b) how volatile each of them is. If they are highly correlated but B is far more volatile than A, you can have a large shortfall if B just has a bad day. If they aren’t very correlated, then when B happens to zig lower as A zags higher, you’ll get a shortfall even if they have similar volatilities. Essentially, we are valuing a spread or Margrabe option and like any option, we need a volatility parameter. In this case, it’s the volatility of the spread we care about, so we can evaluate “what’s the likelihood that the B-A spread is negative.”
If “SA” is the value of an inflation index (or an indexed token like USDi), and “SB” is the value of the hedging asset, then if distributions of A and B are approximately normal,[1] the option value is
C = SA N(d1) – SB N(d2), where
and
and, crucially, is the volatility of the ratio of A to B, which is a formula that will be familiar to travelers in traditional finance and depends on the individual asset volatilities and the correlation () between them:
For this ‘shortfall’ option to be as small as possible, assets A and B should have small volatilities () and a high correlation () between them.
In plain English terms: imagine two drunk guys walking down the boardwalk. What determines how far away they are from each other at any given time? Assuming no drift, it will depend on how much they’re weaving (volatility) and how much they’re weaving in the same pattern (correlation). If they’re holding hands (imposing high correlation), they’ll never get too far away from each other. And if neither one is very drunk (low volatility) they also won’t stray very far from each other. On the other hand, if both are wildly drunk and they don’t know each other, the spread between them will be wildly variable.
We aren’t trying to evaluate the spread between drunks, though. Let’s take this thought process and apply it to the inflation-hedging problem with an example. Suppose you are considering which of two assets is a better ‘hedge’ for inflation: the “INFL” ETF, or a mystery fund – let’s call it “EUSIT.”[2] Here is relevant data for these two assets, and for CPI. These are 3-year returns, volatilities, and month/month correlations, ending November 2025:
Using this data, we can see that the spread volatility (the result of the last formula listed above) for INFL versus CPI is 15.2%, while the spread volatility for EUSIT vs CPI is 1.1%. The Mystery Private Fund is the drunk holding hands with the other drunk, with neither of them that drunk; but INFL is really smashed (14.9% vol) and tending to zig when the CPI drunk zags (negative correlation).
Let’s extend this out one year, assume that INFL, EUSIT, and CPI all have the same expected returns, volatilities, and correlations. Practical question: What is the probability that your investment in INFL or EUSIT underperforms inflation?
For INFL: based on prior returns, it is expected to outperform CPI by 8.99% (11.97% – 2.98%). With a spread volatility of 15.2%, underperforming inflation (a spread of 0% or less) would mean an outcome that is 0.59 standard deviations below the mean. The probability of a draw from a normal distribution being 0.59 standard deviations below the mean is about 33.5%, which means that if you hedge your inflation exposure with INFL, you’ll underperform inflation about one year in three. Your chances of underperforming inflation by 10% or more in a given year is about 18%.
For EUSIT: based on prior returns, it is expected to outperform CPI by 3.15% (6.13% – 2.98%). With a spread volatility of 1.1%, underperforming inflation (a spread of 0% or less) would mean an outcome that is 2.86 standard deviations below the mean. The probability of a draw from a normal distribution being 2.86 standard deviations below the mean is about 0.66%, which means that if you hedge your inflation exposure with EUSIT, you’ll underperform inflation for a full year about once every 151 years. Your chances of underperforming inflation by 10%…even by 5% for that matter…is essentially zero.
Put a star by this paragraph: the assumptions here are key and I am making no claims about either of these strategies having those same characteristics going forward. This is only to illustrate the point that if you want an inflation hedge, meaning that you want to minimize shortfall risk, then it is very important to look at the volatility and correlation to CPI of your intended hedge. Having a better return is important, but less important than you think it is: at a 5-year horizon, the INFL ETF would be expected to outperform inflation (if we think 12% and 3% are decent long-term projections too) by about 60% compounded, but the spread standard deviation is now 15.2% times the square root of 5 years, so you’re only about 1.76 standard deviations above zero and thus you still have an 8% chance of underperforming inflation at the 5-year horizon! On the other hand, your chance of outperforming inflation by a huge amount, if you use the Mystery Fund, is also very small while that possibility exists if you use INFL. That’s what a hedge does: you give up the possibility of big outperformance to ‘buy back’ the chance of underperformance. If you are risk averse, that is a good trade because you’re giving up the less-salient part of your gains (big outperformance) to protect against the more-salient part (big underperformance).
So getting back to answering the question that we started with: what does a good inflation hedge look like?
- It has highly positive correlation to inflation at whatever horizon you’re focused on
- It has low volatility
- It outperforms, or at least doesn’t underperform, inflation over time
To this, I’ll add a fourth characteristic. It’s almost humorous, because hedges that fit those three characteristics are themselves quite rare. But the fourth one I would add is that it has convexity to higher inflation; that is, it does better at an increasing rate, the higher inflation gets. An inflation option, in other words.
Most of us should be happy with three! But at least now you’ll know how to evaluate whether you’re really getting a hedge, or something that will hopefully perform so well that you won’t care that it isn’t a hedge.
[1] I also conveniently wave away some complexities like the relative growth rates and the time value of money to make the math clearer with respect to volatility and correlation, which is my point here.
[2] Mystery fund is a private 3(c)1 fund available to verified accredited investors via a subscription agreement.
