In our previous report, we introduced several common techniques for estimating output gaps in China. In this section, we have updated these estimates to show the latest trend (Exhibit 1):
a. Hodrick-Prescott filter – The most common and best-known technique for estimating output gaps is the Hodrick-Prescott filter. Since intuitively the potential output level only evolves smoothly with time, the HP filter decomposes actual output into trend and cycle components, with trend being the estimated potential output, and cycle being the difference between actual output and potential output, i.e. the output gap. The benefit of this technique is its simplicity. However, there are significant limitations – estimated trend and cycle components can be greatly revised when a new data point is added to the analysis (“endpoint problem”). These drawbacks are also widely discussed in academic papers.1.
b. Christiano-Fitzgerald (CF) filter – Another frequently-used technique for estimating output gap is CF filter. This is a band pass filter which screens out high and low frequency movements in the data series. CF filter then identifies trend, cycle and noise based on frequencies. Similar to other filters, end-point remains an important drawback to this approach.
c. Multivariate state-space model – This model is based on the Philips curve, and explicitly models output gaps with inflation (we used China CPI core inflation) and the lag of core inflation (implying adaptive expectations).2. The drawback of this model is that there is no consensus view on the optimal specification of the Philips curve, and thus estimating output gap based on the Philips curve can be biased.
d. Structural VAR – This follows Blanchard and Quah’s approach3. of decomposing output into supply and demand shocks. Demand shocks impact both output and inflation, while supply shocks impact output but not inflation in the long run. The output gap series is then reconstructed from the demand shocks. This approach has the advantage of being free from revisions (unless of course the underlying output or inflation series are revised).

These four different techniques all provide “top-down” estimates on output gaps.
Among them, we have a modest preference for the structural VAR model which is free from revision, and contains relatively less noise compared with the state-space model.
These top down approaches show that China’s overall output gap remains negative (i.e. actual output is still below potential output) since 2016 at least. However, the majority
also show that the negative output gap has narrowed marginally since late 2016.

Another approach to estimating spare capacity is to look from the “bottom up” – to estimate sector-level capacity utilization measures and then aggregate to an economy-wide measure. In the next part of this report, we develop estimates on capacity utilization rates in various parts of the economy, to supplement the top-down output gap estimates in the first section. We look at capacity utilization rates in
manufacturing, construction, and service sectors.

Slack/excess capacity remains in the industrial sector
We start from the industrial sector. Common industrial capacity utilization measures in other countries are normally based on surveys on goods-producing plants.

In China there are also surveyed equipment capacity utilization rates. NBS has been conducting the survey in the industrial sector over a long period of time, but they don’t regularly release the results4.. PBOC also has capacity utilization surveys on 5,000 key industrial enterprises, although it is a qualitative measure in the form of a diffusion index, and with a significant lag (at this point, Q3 2016 index is the latest available).

With timely official data on industrial capacity utilization rates not yet available in China, we proxied equipment capacity utilization with a few measures below. By definition, capacity utilization can be interpreted as the actual output, divided by the maximum possible output level – or for capital, utilized equipment hours , divided by the overall capacity.

We proxied the maximum level of output with real fixed asset stock in the industrial sector. An implicit assumption here is we assume the maximum output level is
determined by/proportional to the real capital stock (equipment) in the industrial sector.5.

For the actual output level, we tried a few different proxies:

1、Industrial production, which is the real value-added index for the 1. entire industrial sector. The benefit of using industrial production series is that it covers the entire industrial sector, but similar to the official GDP figures, there are also questions around the over-smoothing of IP data. Given China IP is based on value-added, the change of value-added ratio in total industrial output could further affect this estimate.

2. Power consumption by the entire industrial sector. The implicit assumption here is that power consumption will be proportional to actual output, and therefore usingtotal power consumption of the industrial sector is a proxy for total actual output of the industrial sector. The drawback is the coefficient of power consumption to actual output is not constant over time. In fact as our utilities team has discussed in its
research, energy intensity in the industrial sector has been declining by approximately 3% per year since 2000, so we de-trended the power consumption measure to adjust for this.
3. Physical units of production. We used our sales weighted physical output index based on the physical output (number of cars, tons of steel, etc.) in around 30 sub-sectors of the industrial sector.6. The benefit of using physical output is that it’s a direct measure of output. The drawback is that our physical output index is based on heavy industries output. The derived capacity utilization rate is therefore more like the utilization rate of heavy industry.

For each of the three measures listed above, we first calculate its ratio to the real fixed asset stock in each period; and then identify a base period utilization rate. We chose the base period utilization rate as the weighted average utilization rates in auto manufacturing, steel, coal, aluminum, cement, and power sectors, in the year of 2012, at 77%. Utilization rates in these sectors are either published by NBS directly, or estimated by our sector equity research team. We chose 2012 data as the base,  whenthe weighted average of capacity utilization rate in these sectors  was the closest to the NBS irregular data on overall industrial sector utilization rate.

The capacity utilization measures we constructed in the manufacturing sector point to a declining capacity utilization rate over most of the past five years. Three measures of capacity utilization rates in heavy/overall industrial sectors rebounded in Q3/Q4 of 2016.
One point worth noting is that even with the ongoing supply-side reforms, current equipment capacity utilization ratio (by our estimate) is still lower than the level in 2012.
This is also consistent with what the NBS capacity utilization rate series shows. 7.

Comparing with the filter-based output gap estimates
To put our estimated capacity utilization rate into context of output gaps, we briefly compare capacity utilization rates against top-down filter approaches measuring output gaps in the industrial sector only. (Exhibit 3)8.
Filter-based estimates in the industrial sector imply small positive output gaps from 2012 to late 2014. On the other hand, estimated capacity utilization rates had been trending down during that period. In 2016, output gaps narrowed marginally, but capacity utilization measures showed a more visible increase.

excess capacity may not be “effective” anymore, e.g. very old machines lacking maintenance – theoretically, this part should have been removed from the overall capacity estimation in our measures, but in reality it’s hard to quantify the exact amount;

3) “Supply-side reform” measures may impact capacity utilization rate and top-down output gap estimates differently. To the extent “excess capacity” is permanently shut,
this should be reflected in bottom-up utilization data, but would not be reflected in top-down approaches (at least not right away).

Simple Philips curve regressions of secondary industry GDP deflator on output gaps/capacity utilization rate estimates showed that the power consumption estimate seem to have the most predictive power on inflation among the three measures of capacity utilization rate (though the difference among these three measures is small), however, these bottom up capacity utilization rate estimates do not always outperform top down filter-based output gap approaches, especially the HP filter-based output gap. 9.

Using PPI inflation gives similar results, given the trends of PPI inflation and secondary industry GDP deflator are very close.

1.【获得报告】扫一扫下面二维码给群主点赞，同时转发朋友圈， 写下留言或私信，就可获得文中提到的各种研究报告全文链接。

2.【加入讨论】著名券商PE/VC已经加入。扫一扫下面二维码给群主点赞，并转发朋友圈，同时在留言中留下实名制信息（姓名+公司+手机+微信+研究方向），审核通过可加入宏观讨论群，交流各种研究报告和成果。

 3.【点一点最简单】想每天看到新的报告！最简单支持群猪的方法：点一下文章最后的广告！走过路过不要错过，点一点对你也没有损失！