This is an auxiliary article to -"How Made-in-China 2025 is adapting to the US-China Trade War"- which explains the methodologies of predicting industrial policy documents.
GOVERNMENT INTERVENTION: July 2018, the nomenclature "Made in China 2025" was dropped completely from all policy documents. This sounds like a job for data scientists! It just so happens that, JULY 2018 was also the start of the U.S.-China trade war. This is no accident.
RESEARCH QUESTIONS: Considering the name-drop, are the actual themes and objectives of MIC25 completely gone too? If it remains, then have policy priorities changed? and to what degree? (below lists the objectives of MIC25 from 2015)
Step 0: Understand the original MIC 2025 plan, which clearly lists the main themes (or objectives) in 4 topical dimensions of manufacturing upgrade -- 1) innovation advancement, 2) product quality improvement, 3) manufacturing digitization, and 4) green sustainable development. These dimensions are imperative to the understanding of MIC25’s fluctuating dynamics and industrial priorities.
Step 1: TRAINING DATASET -- Gather All the MIC25-Related Policy Documents (PRIOR TO JULY 2018 Intervention) from the Ministry of Industrial and Information Technology and label them as "1". I then randomly sample 500 industrial documents that are not related to MIC25 and label them as "0". TEST DATASET: gather all policy documents published AFTER JULY 2018. The machine learning model will generalize on all the documents published after 2018 to detect any signs of MIC25 still lingering on.
Step 2: Extracting Themes from the MIC25 Policy Documents. The computer-generated model output illustrates there are four major word clusters that are associated with the four major themes of MIC25. Each group of words is paired up with a correlation coefficient to each of MIC25’s specialized themes mentioned earlier (unsupervised learning with LDA algorithm, Policybot.io Analytics, 2019).
Step 3. Feature Engineering. This time instead of using only 4 clusters (matching the 4 major themes) in LDA, I engineered 100 clusters, which I merged with the original dataset in preparation for the model training.
Step 4. Training the Model. I separated the dataset (523 MIC 2025 policy documents) into a 70% training set and 30% validation set. After applying a base 'Lite Gradient Boost' tree-based algorithm to the training set, I rendered an F-score of 97.4% accuracy on the validation set. (I am not going to comment on the model tunning and overfit analysis in this post.) After scoring on the TEST DATASET (which contains no 'mic25' labels), only documents scored more than 90% accuracy are then put back into the analysis. In total there were about 25 additional documents that displayed 'MIC25 themes' after JULY 2018. When additional documents were added back to the dataset, I noticed that there is a change of topical distribution RIGHT AFTER the trade war started (shown below):
RESULTS:
SO I did more investigation...
Post-intervention trends show a sizable change in the allocation of topics. For example, the topics discussed in MIC25 policy documents post-intervention have been realigned to 48% ‘innovation capacity,’ 38% ‘green development,’ 9% ‘quality improvement,’ and 5% ‘digitalization.’ This shows that priorities now are changing to focus more on innovation-building projects and less on digitalization projects.
Tracking the policy flow all the way downstream...
Feel free to ask questions and please leave your comments below if you like this analysis!!!
Disclaimer: This article should not be copied, distributed, published or reproduced in whole or in-part, or disclosed by any recipient to any other person without the permission of the author.
The information contained in this article does not constitute any recommendation from the author. Neither Policybot.io nor the author makes any warranty or representation, expressed or implied, as to the accuracy or the completeness of the statements or any information contained to this article, and any liability thus including, direct or indirect, consequential financial damage or investment loss is explicitly and expressively disclaimed. This article does not provide any economic, financial, or legal recommendations. Thank you for your kind consideration.
