An underlying inflation gauge (UIG) for China

The People’s Bank of China

Abstract

Inflation in China is driven by volatile yet persistent changes in food and energy
prices, making it difficult for policymakers and investors to gauge the underlying
inflation trend. Traditional core inflation measures either neglect or down-weight the
more volatile subcomponents of the CPI and thus risk excluding information that is
helpful in gauging current and future trends in inflation. Therefore, economists at the
PBoC and BIS have jointly explored and developed a novel underlying inflation gauge
(UIG) for China, to complement the traditional core inflation measures. By extracting
the persistent part of the common component in a broad data set of price and non-
price variables, the UIG avoids the excess volatility reduction that plagues traditional
core measures in China’s case. Further, the UIG outperforms traditional core inflation
measures in forecasting the headline CPI over different samples.
Keywords: Inflation, dynamic factor models, core inflation, monetary policy,
forecasting, China
JEL classification: C13, C33, C43, E31, E37, G15

BIS Papers No 89 117

Inflation in China is driven by volatile yet persistent changes in food and energy
prices, making it difficult for policymakers and investors to gauge the underlying
inflation trend. Traditional core inflation measures either neglect or down-weight the
more volatile subcomponents of the CPI and thus risk excluding information that is
helpful in gauging current and future trends in inflation. Therefore, economists at the
PBoC and BIS have jointly explored and developed a novel underlying inflation gauge
(UIG) for China, to complement the traditional core inflation measures. By extracting
the persistent part of the common component in a broad dataset of price and non-
price variables, the UIG avoids the excess volatility reduction that plagues traditional
core measures in China’s case. Further, the UIG outperforms traditional core inflation
measures in forecasting the headline CPI over different samples.

Chinese inflation dynamics

China’s inflation has become lower and less volatile over the past two decades
(Graph 1). Between 1987 and 2000, the mean and standard deviation of the monthly
year-on-year inflation were 8.8% and 8.7%, respectively. During 2001 and June 2012,
however, they dropped to 2.5% and 2.4%, respectively. In this latter period, the
Chinese economy experienced three full “well behaved” inflation cycles. Inflation in
these three post-2000 cycles was much lower and less volatile than the inflation cycles
seen in the 1980s and 1990s. Moreover, China’s post-2000 inflation dynamics appear
to be more associated with domestic and external cyclical shocks and less related to
liberalisation of administered prices.

China’s CPI inflation dynamics (1987–2013, monthly and year-on-year) Graph 1

118 BIS Papers No 89

 Underlying trend in Chinese inflation

Current and prospective inflation matters a lot to monetary policymakers and market
participants. Yet headline inflation can be excessively noisy, making it difficult to
judge whether a sudden up or down move in the most recent CPI observation should
be considered as temporary noise or a change in trend. This has led to the
development of core inflation measures, which either fully exclude or down-weight
volatile subcomponents of the CPI, such as energy and food prices. However, in the
case of China, and in most other emerging markets, food and energy account for a
heavy weighting and play an important role in inflation dynamics. Excluding these
components may lead to an excess volatility reduction. As Graph 2 shows, from 2004
to 2008, China’s official core inflation measures such as the CPI excluding food
(CPI_nf) and the CPI excluding food and energy (CPI_nfe) stayed close to around 1%,
while the CPI itself fell from 5% to 1% and then moved back up to 8%. After getting
rid of these volatile but important components, the traditional core measures show
little movement, and their usefulness in signalling changes in the CPI is thus limited.

Headline and core inflation measures and underlying inflation gauge for China Graph 2

Therefore, economists at the PBoC and BIS have jointly explored and constructed
a novel underlying inflation gauge for China (Amstad, Ye and Ma (2014, 2015)). This
“Underlying Inflation Gauge (UIG) for China” adds instead of discards information that
potentially could be useful to gauge the trend of inflation. The UIG is essentially an
indicator that summarises a broad data set of price, real activity and financial variables
that potentially matter for future inflation but retains only the persistent part of such

BIS Papers No 89 119

information. The data set starts in 2001, has a total of 473 variables and consists of
five major categories: prices (48%), economic activity (35%), the labour market (9%),
money and credit (8%), and the financial market (6%). As these variables are updated
regularly and throughout the month, the UIG can also be updated regularly and intra-
monthly. This feature is particularly useful in times of quick but persistent changes in
inflation – the times when policymakers most need guidance. A version of the UIG
based on a narrower data set of CPI subcomponents only (UIG_ponly) is also
considered.
Technically, the UIG is an application of a dynamic factor model based on Forni
et al (2000, 2005). The model summarises the lead and lag relationship of all included
variables against inflation in the so-called common component. Stock and Watson
(1999) have popularised dynamic factor models. The model type used for UIG retains
only the persistent part in the common component, to assure a smooth and non-
noisy signal.
This approach has already been applied at a number of the OECD central banks,
such as Cristadoro et al (2001) for European inflation and Amstad and Fischer (2009)
at the Swiss National Bank. An underlying inflation gauge for the United States has
been calculated daily at the Federal Reserve of New York since 2005 (Amstad and
Potter (2009), Amstad et al (2016)).
The UIG for China outperforms traditional inflation core measures on several
matrices. While the UIG is highly correlated with and smoother than headline CPI
inflation, it does not suffer from the excess volatility reduction when compared to
core measures. As measured by standard deviation, its volatility of core measures is
only 40–50% of that in the headline inflation, while the UIG retains 83% (Table 1).
More importantly, the UIG forecasts headline inflation better than traditional core
measures, with high statistical significance (Table 2). These results have been robust
in classical forecasting exercises (horse races), over different sample horizons and
across various partitions of the broad data set. Taken together, this illustrates that a
change in the UIG is not due to noise but can be interpreted as a signal that the
underlying trend in inflation is changing. So, while traditional core measures are easy
to construct and communicate, it seems advantageous for policymakers to also follow
the complementary UIG that could provide an early, non-noisy signal, especially when
inflation dynamics are about to change.

Standard deviation
Sample: January 2001–December 2013 Table 1

CPI UIG UIG_ponly UCPI CPI_nf CPI_nfe

S.D. 2.34 1.95 1.81 1.47 1.12 0.95
Portion (%) 100% 83% 77% 63% 48% 40%
Note: S.D. is Standard Deviation. CPI_nf = CPI excluding food. CPI_nfe = CPI excluding food and energy. UIG_ponly = UIG using only price
data. Source: Amstad et al (2014, 2015).

120 BIS Papers No 89

Forecasting performance Table 2

Forecasting full period Forecasting crisis period

2006–13 2008–13
(estimation period:2001–05) (estimation period:2001–07)
DM
RMSE1 DM stat2 RMSE1 DM stat2 DM p-value3
p-value3
UIG 2.63 Na na 2.83 Na na
UIG_ponly 3.03 2.40 0.01 3.15 1.76 0.04
CPI_nf 3.43 2.70 0.00 3.33 2.91 0.00
CPI_nfe 3.29 1.88 0.03 3.10 1.18 0.12
UCPI 3.74 3.17 0.00 3.59 3.09 0.00
CPI_LAG12 4.11 3.41 0.00 4.28 2.59 0.00
1 2 3
Root Mean Square Errors (RMSE). Diebold-Mariano (DM) statistics. Diebold-Mariano likelihood (DM p-value).

Note: CPI_nf = CPI excluding food. CPI_nfe = CPI excluding food and energy. UIG_ponly = UIG using only price data.

Source: Amstad et al (2014, 2015).

References

Amstad M and A Fischer (2009): ”Are weekly inflation forecasts informative?”, Oxford
Bulletin of Economics and Statistics, vol 71, no 2(04).
Amstad, M and S. Potter (2009): “Real time underlying inflation gauges for policy
makers”, FRBNY Staff Report, no 420.
Amstad, M, S Potter and R Rich (2016): “The FRBNY staff Underlying Inflation Gauge
(UIG)”, Economic Policy Review (forthcoming).
Amstad, M, H Ye and G Ma (2014): “Developing an underlying inflation gauge for
China”, BIS Working Papers, no 465.
——— (2015): “对中国基础通货膨胀指标的研究 – Developing an underlying inflation
gauge for China”, PBC Working Paper, no 2015(5).
Cristadoro, R, M Forni, L Reichlin and G Veronese (2001): “A core inflation index for
the euro area”, Journal of Money, Credit and Banking, vol 37, no 3.
Forni, M, M Hallin, M Lippi and L Reichlin (2000): “The generalized factor model:
identification and estimation”, The Review of Economics and Statistics, vol 82,
pp 540–54.
——— (2005): “The generalized dynamic factor model: one-sided estimation and
forecasting”, Journal of the American Statistical Association, vol 100, pp 830–40.
Stock, J, H and M Watson (1999): “Forecasting inflation,” Journal of Monetary
Economics, vol 44, no 2, pp 293–335.

BIS Papers No 89 121