First I plot graphs for real GDP per capita and the Human Development Index for three countries: US, China, and Norway (the highest ranked in 2019 by the HDI).

**Figure 1:** US real GDP per capita, in PPP Chained 2017 US$ (blue, left scale), and Human Development Index for US (red square). Source: Penn World Tables 10.0, and UNDP.

**Figure 2:** Chinese real GDP per capita, in PPP Chained 2017 US$ (blue, left scale), and Human Development Index for China (red square). Source: Penn World Tables 10.0, and UNDP.

**Figure 3:** Norwegian real GDP per capita, in PPP Chained 2017 US$ (blue, left scale), and Human Development Index for Norway (red square). Source: Penn World Tables 10.0, and UNDP.

The HDI is briefly described thusly (2020 Technical Notes):

The Human Development Index (HDI) is a summary measure of achievements in three key dimensions of human development: a long and healthy life, access to knowledge and a decent standard of living. The HDI is the geometric mean of normalized indices for each of the three dimensions.

Graphically:

**Source:** UNDP (2020).

Note that per capita GNI (quantitatively similar to GDP usually) is a component of the HDI.

Obviously, the two series don’t comove exactly (as far as we can tell with these data). However, both per capita real GDP and the HDI co-trend. A naive regression of HDI on log per capita GDP for US and China yields a coefficient of about 0.15.

**Figure 4:** US real GDP per capita, in PPP Chained 2017 US$ (blue), Chinese (red), and Norwegian (green). Source: Penn World Tables 10.0.

**Figure 5:** US Human Development Index (blue squares), Chinese (red squares), and Norwegian (green squares). Source: UNDP.

So, now one knows – a broader measure of economic welfare typically co-trends, but there are surely deviations of one from the other; and cross-country comparisons of GDP per capita (even in PPP terms) and economic welfare will probably differ in terms of ranking.

I would not assert HDI is the best indicator; I’m only using it because it’s easily available and has been around for a long time. As the technical notes to the HDI lay out, there are additional versions, that account for income inequality and gender inequality, or focus on poverty.

There are innumerable papers on the subject of alternative means of measuring economic welfare. Rather than trying to cite most of them, I’ll refer the reader to one recent review (IMF, 2020) [link corrected 1pm]. See also this review article by Dale Jorgensen, “Production and Welfare: Progress in Economic Measurement” (JEL, 2018).

]]>

**Figure 1:** GDP (black), implied GDP from Atlanta Fed 10/19 nowcast (red triangle), 10/27 nowcast (dark red triangle), IHS-MarkIt (sky blue square), Bloomberg consensus (brown square), Goldman Sachs 10/26 tracking forecast (inverted pink triangle) 10/27 tracking forecast (inverted dark pink triangle), and mean forecast from WSJ October survey (green line). Levels calculated using reported growth rates and latest GDP for Q2. Source: Atlanta Fed, IHS-Markit, Bloomberg, Goldman-Sachs, WSJ October survey, and author’s calculations.

The Bloomberg consensus of 10/26 was not changed, while the IHS-MarkIt nowcast rose from 1.5% to 1.6% q/q SAAR – small enough I did not plot the new implied level.

]]>

**Figure 1:** Three month Treasury yields on secondary market, monthly average of daily data (black), Survey of Professional Forecasters mean forecasts from indicated quarters. NBER recession dates shaded gray. Source: Federal Reserve and Philadelphia Fed Survey of Professional Forecasters, and NBER.

To a certain extent, the tendency to project falling rates when interest rates are high, and to project rising rates when they are low, is a reflection of the belief in mean reversion. On the other hand it’s interesting that even at turning points, consensus views wildly miss the trajectory of short rates.

For forecasts of ten year Treasury yields, see this post, where a similar pattern is found. However, the 3 month Treasury is closely linked to the Fed funds rate (see Rudebusch for those forecasts), so this outcome is really in a way a reflection of how poorly economists can figure out the Fed’s reaction to developments that themselves are largely unpredictable (meltdown in shadow banking, the Covid pandemic).

]]>

**Figure 1:** GDP (black), implied GDP from Atlanta Fed (10/19) (red triangle), IHS-MarkIt (sky blue square), Bloomberg consensus (brown square), Goldman Sachs (inverted pink triangle), and mean forecast from WSJ October survey (green line). Levels calculated using reported growth rates and latest GDP for Q2. Source: Atlanta Fed, IHS-Markit, Bloomberg, Goldman-Sachs, WSJ October survey, and author’s calculations.

The Bloomberg consensus reported as of today on the website is pretty much at the WSJ survey mean taken in the first week or so of October. As for nowcasts, the Atlanta Fed’s GDPNow estimate is below the IHS-MarkIt (nee Macroeconomic Advisers) nowcast and substantially below the Goldman Sachs tracking estimate. However, the GDPNow nowcast is now a week old, with a new release set for tomorrow.

A final thought: Even if the advance release reports 0.5% q/q growth SAAR, the mean absolute revision going from advance to second, and going from advance to third release are, respectively, 0.53 and 0.59 percentage points (1996-2018) (see here for more documentation).

]]>

**Figure 1:** Five year inflation breakeven calculated as five year Treasury yield minus five year TIPS yield (blue), five year breakeven adjusted by inflation risk premium and liquidity premium per DKW (red), all in %. NBER defined recession dates shaded gray (from beginning of peak month to end of trough month). Source: FRB via FRED, Treasury, KWW following D’amico, Kim and Wei (DKW) accessed 10/7, NBER and author’s calculations.

What to make of this development?

The gap between the unadjusted and adjusted spreads was 0.44 ppts as of 9/30. If that gap — the combination of inflation risk and liquidity premia — had stayed the same, the implied inflation rate would be 2.5% over the next five years. (Of course, there’s no particular reason why the composite premium would have stayed constant…)

Analysts often look at the the five year five year forward as an indicator of inflation over the five years five years from now. If one applies the same logic about premia to the usual calculation, then one obtains Figure 2.

**Figure 2:** Five year five year forward spread (blue), five year five year forward spread calculated using inflation risk premium and liquidity premium per DKW (red), all in %. NBER defined recession dates shaded gray (from beginning of peak month to end of trough month). Source: FRB via FRED, Treasury, KWW following D’amico, Kim and Wei (DKW) accessed 10/7, NBER and author’s calculations.

Two observations:

- The “adjusted” spread requires estimating inflation risk and liquidity premia, so one might worry about all sorts of errors being introduced. For what it’s worth, the adjusted spread forecasts actual inflation better than the unadjusted does (although the sample is small), as discussed in this post with an exchange with Joseph E. Gagnon and Madi Sarsenbayev of the Peterson Institute.
- The spread predicts CPI inflation. The Fed pays attention to the PCE deflator. Over the last 35 years, CPI inflation has averaged about 0.43 ppts more than PCE inflation.

If one believed in the DKW adjustment (and that adjustment held constant over the past four weeks), and the gap between CPI and PCE inflation remains roughly constant, that would place average PCE inflation over the next five years at about … 2%.

The five year five year forward, which is often taken as long term CPI inflation, is then 2.3%; implied PCE inflation is then 1.9%…

]]>

**Figure 1:** Price of WTI oil, deflated by Core CPI (blue), price of oil (10/25) deflated by nowcast Core CPI (sky blue square). NBER defined recession dates shaded gray. Source: FRED, Cleveland Fed, NBER and author’s calculations.

The September 24 1982-84$/bbl price of 26 is a little over the 1967-2021 mean of 21; the October 25th guesstimated relative price is 30, still less than a standard deviation above mean.

On the other hand, in terms of shocks, one might be interested in the change in relative prices within a short period of time. Figure depicts the year-on-year growth rates (calculated in log-differences).

**Figure 2:** Growth rate (y/y) of price of WTI oil, deflated by Core CPI (blue), growth rate of price of oil (10/25) deflated by nowcast Core CPI (sky blue square). NBER defined recession dates shaded gray. Source: FRED, Cleveland Fed, NBER and author’s calculations.

The mean y/y growth rate is 0.016, while September’s value is 0.553. The standard deviation of changes in 0.323, so September’s change is about 1.7 standard deviations from mean. If October average price matches today’s price ($83.60 when last I checked on 10/25), then the y/y growth rate will be 0.710.

]]>

Figure 1 depicts real GDP over the period NBS reports it:

**Figure 1:** Chinese real GDP, in 100 mn 2015 CNY (blue), seasonally adjusted using Census X-12/X-11 seasonal filter (black), and IMF WEO forecasts applied to seasonally adjusted series (sky blue squares), on log scale. GDP in constant 2015CNY calculated by author by using IMF WEO GDP price deflator to convert implicit deflators to a common base year. Source: China National Bureau of Statistics, IMF October 2021 WEO, and author’s calculations.

Using the seasonally adjusted, and applying the IMF’s forecasts to these seasonally adjusted numbers, it’s clear how much the Q3 figure was a miss given expectations.

In my mind, it’s a bit mysterious how the NBS q/q growth numbers were obtained. On the NBS website, seasonally unadjusted nominal GDP numbers are reported, as are seasonally unadjusted real numbers in *different* constant yuan, and seasonally adjusted q/q real growth rates (from 2011 onward). That means it is not straightforward to replicate the implied Chinese seasonally adjusted GDP level, since they use an in-house seasonal adjustment methodology (NBS-SA, see here). I apply several seasonal adjustment procedures, including a geometric ratio to a moving average, Census X-12/seasonal filter X-11, and ARIMA Census X-12/seasonal filter X-11, and compare against the implied GDP series obtained by cumulating NBS reported q/q changes.

**Figure 2:** Chinese real GDP estimated using Census X-12/seasonal X-11 (black), ARIMA Census X-12/seasonal X-11 (sky blue), ratio to moving average (green), and implied by NBS reported q/q growth rates (red), all in logs, normalized to 2019Q1. *Source: NBS, and author’s calculations.*

Notice that the official series indicates continued growth albeit slow, while two of three standard seasonal adjustment procedures show negative growth. This can be seen in the graph of growth rates (calculated as log-differences):

**Figure 3:** Quarter-on-quarter growth rates of Chinese real GDP estimated using Census X-12/seasonal X-11 (black), ARIMA Census X-12/seasonal X-11 (sky blue), ratio to moving average (green), and NBS reported q/q growth rates (red), all calculated as log-differences. *Source: NBS, and author’s calculations.*

Over the last two quarters, cumulative growth estimated using X-12 adjusted GDP has lagged NBS reported growth, raising the question whether the numbers have been massaged. Some people would respond that it’s a foregone conclusion that they were. However, as noted in this post, Chinese measures of economic activity are not obviously over- or understating actual activity, particularly in recent years. Fernald, Hsu and Spiegel (2020) [working paper version] construct a “China Cyclical Activity Tracker” (C-CAT) [see letter], and conclude:

We choose a preferred index of eight non-GDP indicators based on their fit to Chinese imports, which we call the China Cyclical Activity Tracker (or C-CAT). We find that Chinese statistics have broadly become more reliable in measuring cyclical fluctuations over time. However, measured GDP has been excessively smooth since 2013, and adds little information relative to combinations of other indicators.

The authors have done a preliminary check on the 2021Q3 GDP, using a new, unpublished *quarterly* version of their C-CAT (the original published version is on a 4 quarter basis). They note that using Haver data — which does not match NBS data –, reported GDP is 31 basis points (bps) below trend in standard deviation units, while their preliminary reading indicates 119 bps below trend for a preliminary reading of *q-on-q* estimates of the China CAT and 170 bps below trend for q-on-q estimates of a quarterly version of their “All indicators” index (both without consumer expectations, which have yet to come in for the quarter). Those preliminary results confirm the reported results in showing a slowdown for Q3, but look even a little weaker than even what non-NBS GDP series report. (Thanks to Fernald and Spiegel for their results).

(Haver’s series indicates -2.26% q/q growth SAAR, well below NBS’s -0.8%, and closer to my X-12 series (black line) above, -2.3%).

Journalistic accounts (e.g., NYT) have attributed the Q3 slowdown to the financial fallout from the Evergrande default as well as more general real estate troubles, slowed auto production due to chip shortages, and power outages/shortages.

The deceleration is important for the obvious reason that Chinese GDP now (2021) accounts for a larger share of world GDP than it did back in 2008: 17.8% vs. 7.2%; the 2021 share for the US is 24.2% (all shares calculated using GDP in US dollars at market exchange rates as reported in the IMF’s WEO database).

What’s the spillover effect? Ahmed et al. (2019) examine implications of a China slowdown (discussed in this post).

…the VAR estimates also suggest that the hit to economic activity in different countries and regions would generally be significant (consistent with China’s strong trade links with other economies). More specifically, the output hit to EME commodity exporters would be about ¾ as large as the hit to China itself; to other EMEs would be about half; to advanced economies excluding the United States slightly more than a third; and only a relatively modest hit to the United States. The smaller U.S. effect reflects the U.S. economy being more closed, limited direct U.S. financial linkages to China, and greater capacity at the moment (than other advanced economies, say) to ease monetary policy to cushion the blow.

The impact of a 4% shock (blue), estimated using a SVAR, is shown below:

**Source:** Ahmed et al. (2019).

Of course, the relationships that obtained pre-pandemic are unlikely to hold fully in the current environment, but the implications are straightforward. The impact also depends on the persistence of the shock; the more long-lived the measures implemented by the Xi regime, the less likely the resumption of rapid growth (at least in the short term). In a recent op-ed, Arthur Kroeber lays out the macro challenges facing China in the context of Xi’s objectives, and hence the motivation for these regulatory measures.

]]>

This assertion is based upon the following graph of futures:

**Source:** Pescatori, Steurmer, Valckx (2021).

Futures are pointing downward. How much faith should we put into these market indicators? Remember, only under the assumption of risk neutrality should the basis be an unbiased predictor of future spot rates. This characterization is certainly untrue for exchange rates, and precious metals. For energy, Chinn and Coibion (JFM, 2014) assess the data from 1990-2012.

First, a regression of the change in spot rate on the basis. Table 1 shows the results for all commodities examined, with yellow highlights for energy commodities. In general, there are no rejections of the null hypothesis of β = 1 for petroleum, and only one for natural gas.

**Source:** Chinn and Coibion (2014).

Admittedly, R^{2}‘s are pretty low for petroleum, less so for natural gas.

What about prediction? Table II addresses that issue. Once again, I highlight in yellow energy futures.

**Source:** Chinn and Coibion (2014).

In terms of *prediction*, futures outperform a random walk, correctly predict direction-of-change more often than random chance, with statistical significance. Out of sample, futures don’t do any worse or better than a random walk. A random walk outperforms any time-series model that has coefficients to be estimated.

Table 1 reports full-sample estimates. The degree of bias exhibits time variation, as shown in Panel A of Figure 5.

**Source:** Chinn and Coibion (2014), Figure 5, Panel A. **Note:** Each ﬁgure plots estimates of the coefﬁcient on the futures basis from Equation (3) in rolling ﬁve‐year regressions along with 95% conﬁdence intervals (dashed lines) for each commodity and forecasting horizon.

If one is more concerned about the direction of change, Panel B of Figure 7 is interesting. They show the fraction of correct sign predictions relative to 0.5 – so a value of 0.2 indicates energy futures correctly predict price changes more than random chance by 20 percentage points.

**Source:** Chinn and Coibion (2014), Figure 7, Panel B. **Note:** Panel B plots ﬁve‐year rolling fractions of correct sign predictions (using ﬁrst‐differences of basis) minus their unconditional expectation, averaged across 3‐,6‐, and 12‐month horizons for all commodities within each commodity group.

Hence, energy futures have informational content, but declining over the sample period (ending 2012).

One caveat: the futures evaluated are mostly US (e.g., natural gas is for Henry Hub), so I’m not presenting empirical results for European natural gas.

In sum, the fact that energy futures are pointing downward is reassuring. However, one should not rely too much on the idea that future spot prices will go down as far as implied by futures, nor should one be surprised if actual prices turn out to be very high or very low relative to predicted (given the low proportion of variation explained by the typical regression).

]]>

Hard to know where this is all going to lead. But one thing is clear– we have added a very interesting new chapter in the history of money.

In my course on the financial system, I’ve had to update the material to include cryptocurrencies and central bank digital currencies (CBDC). Here’s some pictures of cryptocurrencies.

**Figure 1:** Price of bitcoin (blue), ethereum (brown), litecoin (green), in USD, in logs, 2017M01=0. NBER defined recession dates shaded gray. Source: FRED, NBER.

These three particular cryptocurrencies have experienced proportionately enormous appreciations. Taking bitcoin as an example, it’s clear cryptocurrency returns have been enormous compared to even the S&P 500.

**Figure 2:** Price of bitcoin in USD (blue), London 3pm price of gold in USD/oz. (brown), S&P 500 index (green), in logs, 2017M01=0. NBER defined recession dates shaded gray. Source: FRED, NBER.

However, the month-on-month volatility of bitcoin is enormous, even dwarfing that of gold, as shown in Figure 3.

**Figure 3:** Month-on-month growth rate of the price of bitcoin in USD (blue), of London 3pm price of gold in USD/oz. (brown), of S&P 500 index (green), all calculated as log-differences. NBER defined recession dates shaded gray. Source: FRED, NBER.

The standard deviation of month-on-month (not annualized) changes was 2.8% and 3.9% for gold and S&P 500 respectively. For bitcoin, it’s 21.6% *monthly*. That means that bitcoin does not fulfill the third function of money, namely a store of value, very well.

Given this volatility, one has to wonder why one would want to hold bitcoin. In his post, Jim asks:

Why does the stuff have value in the first place? The answer is that it would be very helpful to many buyers and sellers of real goods and services if they were able to pay for transactions in this way. We can think of any form of money as an asset that provides liquidity services, which refers to the tangible benefit to its holder coming from the ability of the asset to facilitate certain transactions. The value of money, that is, the value of real stuff you’d be willing to give up to hold money, can be thought of as the present value of the stream of these future liquidity services.

Bitcoin has two potential advantages over credit cards for providing such liquidity services. First, the supporting network only needs to verify that the private code is valid, which is less costly than verifying that you are indeed the rightful owner of a credit card and are ultimately going to deliver good funds. …

Second, Bitcoins are relatively more anonymous than credit cards. In this respect, they enjoy some of the same advantages as cash….

One can formalize this argument by referring to the equation for pricing assets:

*D* stands for dividends when *P* refers to a stock price. In our context, *D* is the liquidity services provided by bitcoin (which can be small for those who don’t need to evade restrictions), *P* the price of bitcoin. If one can rule out bubbles, then a bitcoin price is equal to the present discounted value of liquidity services. However, there’s no reason to impose this assumption.

Then the price of bitcoin is moved primarily by new information that changes the information set used for forecasting the future price — in other words, the speculative motive is central.

The expected future price is in this interpretation driven by new information about the liquidity services provided by bitcoin. New regulatory measures — either tightening or loosening — should be associated with bitcoin price movements. Figure 4 highlights the role of such regulatory events, as well as the discount rate.

**Figure 4:** Price of bitcoin in USD (blue, left log scale), TIPS 5 year yield, in % (brown, right scale). NBER defined recession dates shaded gray. Source: FRED, NBER.

Chinese measures to rein in the use of bitcoin negatively impacted prices. On the other hand increasing acceptance of the use of bitcoin — as in the establishment of a bitcoin futures ETF — enhanced the liquidity services provided by bitcoin.

What does the future herald for the price of bitcoin? It depends on the balance between increasing regulations that limit the desirability of bitcoin as a pseudonymous means of transactions and the increasing usefulness of bitcoin as an asset class. The establishment of central bank digital currencies (CBDCs) will also certainly alter the relative desirability of cryptocurrencies.

For more, see Charles Engel’s paper on the subject. Eswar Prasad devotes considerable discussion of cryptocurrencies in his new, comprehensive assessment of the digital revolution in finance, The Future of Money.

]]>

**Figure 1:** CPI-all urban (blue), and CPI-wage earners and clerical workers (red), s.a., in logs 2020M02=0. NBER defined recession dates shaded gray. Source: BLS, NBER and authors calculations.

**Figure 2:** Year-on-year inflation rates for CPI-all urban (blue), and CPI-wage earners and clerical workers (red), s.a., calculated as log-differences. NBER defined recession dates shaded gray. Source: BLS, NBER and authors calculations.

Inflation for the bundle that wage earners/clerical workers has outpaced that for all-urban, by about 0.6 ppts by September.

Interestingly, the weights for the two CPI bundles indicate that wage earners/clerical workers have a higher weight on food, food away from home, and private transportation, and less weight on housing, than all urban consumers. As elevated housing costs feed into the CPI housing components, the places might switch.

**Update, 6:30pm Pacific:**

Barkley Rosser wonders about a CPI for those over age 62. There is, but it is a research — rather than official — series. See here for discussion of that BLS research series. I show that series, along with the research Harmonized Index of Consumer Prices (HICP) (BLS description here).

**Figure 3:** CPI-all urban (blue), CPI for of 62 (crimson), both s.a., and Harmonized Index of Consumer Prices (HICP), n.s.a., all in logs 2020M02=0. NBER defined recession dates shaded gray. Source: BLS, NBER and authors calculations.

Note that while the HICP includes both rural and urban consumers, it excludes housing costs (and is not seasonally adjusted). In comparing the HICP to the CPI, one would want to use a CPI excluding shelter.

]]>