The WorldofGovernmentBonds website provides 2s10s, 2s5s and 1s2s spreads for a whole bunch of countries (missing unfortunately my favorite the 3m10s). Here’s the yield curve for China as of today:

**Source:** worldofgovernmentbonds.com.

For contrast, here’s the US:

**Source:** worldofgovernmentbonds.com.

It is this flattening of the US yield curve that has sparked so much commentary (see here).

Despite the fact that this site tabulates these spreads such as the 10yr-2yr (aka 2s10s), do we know much about what this means for the countries that have inversions?

**Source:** worldofgovernmentbonds.com.

Well here’s a time series plot of spreads and year-on-year industrial production growth.

**Figure 1:** 10yr-2yr Chinese government bond spread (red, left scale), 10yr-overnight interbank spread (blue, left scale) both in %, and year-on-year industrial production growth, % (black, right scale). ECRI defined recession dates peak-to-trough shaded gray. Source: investing.com, OECD via FRED, ECRI, and author’s calculations.

The spikes at the beginning of the sample in the y/y IP growth are due to the lunar new year holiday moving around between January to February.

If you were hard pressed to visually find a relationship between lagged term spreads and 12 month IP growth, you could be excused. A simple regression yields:

Δ*ip _{t+12}* =

**8.73**+

**2.62**

*spread*

^{10yr-2yr}_{t}Adj-R^{2} = 0.12, SER = 4.80, DW = 0.29, N = 212, where **bold** denotes significance at 5% using HAC robust standard errors.

However, the result is not robust to inclusion of a trend (which is apparent in industrial production growth).

Δ*ip _{t+12}* =

**152.68**+ 0.43

*spread*– 0.06

^{10yr-2yr}_{t}*time*

_{t}Adj-R^{2} = 0.68, SER = 2.88, DW = 0.74, N = 212, where **bold** denotes significance at 5% msl using HAC robust standard errors.

Nor is the result robust to truncating the sample at 2014 (where IP growth stabilizes somewhat).

Δ*ip _{t+12?sub>}* =

**6.10**+ -0.51

*spread*

^{10yr-2yr}_{t}Adj-R^{2} = -0.01, SER = 2.76, DW = 0.51, N = 79, where **bold** denotes significance at 5% msl using HAC robust standard errors.

There’s a lot of reason to believe that the predictive power of the term spread in the case of the US would not transfer to China, given the segmented nature of the Chinese government bond market. As discussed in Chen et al. (2019), the market is highly segmented, with little liquidity until recently at longer maturities like10 years. Most of the transactions were at maturities of 3 years. To my knowledge, no studies assessed the predictive power of long spreads;

Moving to quarterly data:

**Figure 2:** 10yr-2yr Chinese government bond spread (red, left scale), 10yr-overnight interbank spread (blue, left scale) both in %, and year-on-year industrial production growth, % (black, right scale). ECRI defined recession dates peak-to-trough shaded gray. Source: investing.com, OECD via FRED, ECRI, and author’s calculations.

**Figure 3:** 10yr-2yr Chinese government bond spread (red, left scale), 10yr-overnight interbank spread (blue, left scale) both in %, and year-on-year GDP growth, % (black, right scale). Nominal GDP deflated by deflator interpolated from annual series from WEO ECRI defined recession dates peak-to-trough shaded gray. Source: investing.com, NBS, IMF WEO (April 2022), ECRI, and author’s calculations.

Similar fragility of results are obtained using either industrial production or GDP; the results are very sensitive to the inclusion of time trend. The most positive results (in terms of finding a role for the term spread) is:

Δ*y _{t+4t}* =

**7.60**+

**1.43**

*spread*

^{10yr-2yr}_{t}Adj-R^{2} = 0.06, SER = 3.40, DW = 0.78, N = 75, where **bold** denotes significance at 5% msl using HAC robust standard errors.

Where the spread coefficient drops to statistical non-significance with the inclusion of a time trend. Putting in a shift term for 2014-2022:

Δ*y _{t+4t}* =

**9.35**+

__0.71__*spread*–

^{10yr-2yr}_{t}**3.11**dummy

^{2014-2022}

_{t}

Adj-R^{2} = 0.22, SER = 3.09, DW = 0.90, N = 75, where **bold** denotes significance at 5% using HAC robust standard errors, and * italic_underscore *denotes significance at 17% msl using HAC robust standard errors.

The failure to find a systematic relationship similar to those found for the advanced countries (recalling that Chinn and Kucko (*Int.Fin.*, 2015) failed to find it for advanced economies Canada, France, Italy, Japan, UK) is not surprising, given the extant literature (as far as I can find; Mehl (*OER*, 2009) investigates 5yr-3mo for emerging markets, while Haubrich (*Ann.Rev.*, 2021) [ungated 2020 wp version] reviews some of the more recent evidence). While Sowmya and Prasanna (*IREF*, 2018) find a relationship between slope and subsequent output, it’s a negative relationship. Chang, Mattson and Tang (*IJFS*, 2019) find a larger adjusted one year-interbank spread predicts *slower* output growth. Interestingly, Jiang, Guo and Zhang (*Econ.Mod.*, 2017) don’t find a spread as one of the useful predictors for growth.

So, bottom line: you can calculate spreads, but the big question is whether they mean anything (for growth).