Enodo first introduced its nowcast in July 2020 to forecast both official Chinese GDP data and our own estimate of real GDP performance. Nowcasting uses complex econometric techniques and a broad set of contemporaneous data to provide a timely view of economic momentum. 

But the Covid-19 crisis threw a spanner in the works and made us question our model's usefulness.

For the second and fourth quarters of 2022, the nowcast resulted in a significant overestimation of Chinese GDP, prompting us at the start of this year to revisit our model and assess its performance. 

We found that the model performs well in normal periods, but was thrown off by the variety of negative shocks that disrupted the Chinese economy as it imposed severe lockdowns in a last-ditch attempt to fight off Covid.

Enodo Economics remains confident in our monthly nowcast, especially now that China has fully reopened its economy and most economic activity has returned to normal, and we will continue to publish monthly nowcasts in our chart library.

But evaluating China's economic performance in Q1 has always been a challenge due to the difficulty in removing the distortions arising from the varying timing of the Chinese New Year. As such January is a blind month and we place little value in the February vintage of the nowcast. We tend to wait until March for our overall cyclical growth analysis, including the nowcast. 

With this caveat, nonetheless, it is worth noting that our February nowcast shows strong growth in Q1 at 3.2% qoq for the official number and 4.2% qoq for Enodo real GDP growth estimate. 

Annual growth in the first quarter is expected to accelerate to 4.7% and 7.3% respectively. Positive surprises on soft indicators, such as stronger-than-expected PMI numbers and improving consumer confidence, were the main reason behind our nowcast for a robust growth rebound already in the first quarter of this year.  

As we discussed in mid-January, Beijing has opted for a sizable fiscal and monetary boost, which is likely to result in growth rates that come close to pre-pandemic levels, at least in the short term. But state efforts to support the economy throughout 2023 are likely to have diminishing returns as the positive base effect wanes. 

Rather than rely on state-mandated, credit-driven investment spending China must nurture its own consumer demand. 

Beijing is aware and its plans have started to take shape. Outgoing Premier Li Keqiang may sketch out parts of the administration’s strategy further when he delivers the annual Government Work Report to the National People’s Congress next month. Over the next few weeks, we will discuss the policies and prospects in a series of Enodo Insights, starting with "China Attempts to Unlock its Own Internal Market". 

Appendix: Enodo Nowcasting Assessment

We monitor a large set of key data and what they tell us about GDP growth in real-time, employing a dynamic factor model (DFM) to extract latent factors driving movements in the economic data and produce a forecast for each series that we track.

The note presents an evaluation of the dynamic factor model (DFM) employed by Enodo to estimate China's quarter-on-quarter GDP growth rate. The evaluation is based on empirical data spanning ten forecasting rounds, during which the Enodo team generated approximately 3 forecasts per round for GDP growth in the current (Qt) and next quarters (Qt+1). The nowcasting model is executed monthly, resulting in 6 estimates per quarter (represented in the text with subscripts t, t-1, t-2, t-3, t-4, and t-5).

Official GDP data

The chart below displays the "spaghetti" graph of the nowcast estimates. It shows a series of predictions from multiple vintages. A visual inspection of the graph suggests that the model accurately tracks GDP growth, with some exceptions.

During periods of rapid GDP growth deceleration or decline (e.g. Q2 and Q4 2022), the model appears to overestimate economic activity. Conversely, during periods of strong economic recovery (e.g. Q3 2022), the model may underestimate actual output.

This can be attributed to both the model's structure and specific factors in China during the period under review, such as the reintroduction and later phasing out of COVID-related containment measures. The short-term indicators used for the nowcast did not immediately reflect the full economic impact of containment measures, making the model slow to capture abrupt changes in GDP growth.

It's important to note that the model's forecast error is closely tied to the volatility of GDP data, and both should be considered together. From Q2 2020 to Q4 2022, China's average GDP growth rate was 2.2% with a standard deviation of 3.5 percentage points. The coefficient of variation, which is a relative measure of variability and is calculated by dividing the standard deviation by the mean (multiplied by 100), is 160.2% in this case[1].

The mean error, mean absolute error, and root mean squared error are three common measures used to evaluate the accuracy of a forecasting model. We measure the errors for our nowcasting model for the whole period and for the period excluding the two problem quarters of Q2 and Q4 2022. Their values are as follows: 

For the period excluding Q2 and Q4 2022, the mean error is negligible and negative implying that the nowcast tends to underestimate actual GDP growth. All errors are small relative to the large standard deviation of the underlying GDP series, suggesting that our nowcast performs well.

Mean error (ME): ME measures the average deviation of the forecast from the actual value. It is calculated as the difference between the forecast and the actual value, divided by the number of forecasts. ME is a simple measure of accuracy, but it has limitations since it does not consider the magnitude of the error. Positive and negative errors can cancel each other out, which can give a misleading measure of accuracy.

Mean Absolute Error (MAE): MAE is a measure of the average magnitude of the forecast errors. It is calculated as the sum of the absolute differences between the forecast and actual value, divided by the number of forecasts. MAE is a better measure of accuracy than ME as it does not have the limitations of ME and is more sensitive to the magnitude of the error.

Root Mean Squared Error (RMSE): RMSE is a measure of the average magnitude of the forecast error. It is calculated as the square root of the average of the squared differences between the forecast and actual value. RMSE is more sensitive to large errors than MAE, and it is more commonly used to evaluate forecasting models because of its interpretability.

When analyzing the errors in forecasts over time, we have observed that the errors for the current quarter are generally slighter than the errors for the upcoming quarter. On average, the difference between these two is substantial.

Enodo’s estimate of GDP

[1] The standard deviation is a measure of how much the individual observations in a data set deviate from the mean or average of the data set. It provides a quantitative indication of the amount of variation or dispersion of a set of values. The coefficient of variation (CV) is a relative measure of variability that is expressed as a percentage. It is calculated by dividing the standard deviation by the mean and multiplying the result by 100. CV is useful in comparing the variability of different data sets that have different units or scales. By expressing the variability as a percentage of the mean, CV allows for comparisons of variability across data sets with different scales. In short, the standard deviation is a measure of absolute variability, while the coefficient of variation is a measure of relative variability.