How China’s censorship on trade war throws off the algos - Trade Stocks

How China’s censorship on trade war throws off the algos

By Tue, May 14, 2019

When China’s stock-market was rocked by its biggest one-day selloff in three years on May 6, Chinese state media and local bloggers steered clear of the elephant in the room – President Donald Trump’s threat to hike tariffs on Chinese goods.

Quantitative investors say censorship around the chief source of China’s equity tumble highlights the limitations of algorithms used to sift through millions of social media posts and news articles in real time to mine investor sentiment. This comes as quantitative hedge funds and well-resourced money managers, such as BlackRock Inc. BLK, +0.68% advocate looking at social media data to check the pulse of unsophisticated retail traders in China’s stock markets.

The analysis of social media first got its start as part of a broad movement in Wall Street to analyze unconventional data sources such as satellite images and credit-card transactions. In developed markets like the U.S., where trade is largely driven by larger institutional investors, social media was mostly used to root out sources of undiscovered information that had yet to affect stock values.

But Phillip Wool, director of research at Rayliant Global Advisors, said that in China, quants use such data in China to find how much mom-and-pop investors, who often act more on speculation and not fundamentals, overreacted to market-moving events like a product launch or allegations of accounting manipulation.

He said there is plenty of opportunity to seize on investor mistakes when retail traders are responsible for more than 80% of turnover in China-listed stocks, according to the China Securities Regulatory Commission.

To do that, quants might train algorithms to scan Chinese-language blog posts on social networking platforms for the frequency of certain phrases that carry positive or negative implications to infer the overall tone of the writings, and then assess how spikes in the number of bullish or bearish blog posts influence stock prices.

Luo Yin, head of quantitative research at Wolfe Research, said stock pickers found social media algorithms especially effective when aimed at coverage surrounding individual Chinese companies. As these narrower issues attract less censorship, news articles and online discussions around such topics showed a richer variety of opinions and therefore gave a better indication of what local investors thought.

But Luo, the former global head of Deutsche Bank’s quantitative strategy team, said these same algorithms were less effective when they faced broader and more sensitive topics like the U.S.-China trade spat, discussions over which Beijing has attempted to keep a tight rein.

Parsing Chinese news and social media on politically sensitive matters such as the state of play in trade negotiations was less useful and revealing in part because nonofficial media sources and blog posts will toe Beijing’s official line, or risk being censored, said Luo.

“Given the censorship in China, voices tend to be less diverse and more aligned with state media,” he said.

After Trump threatened to raise tariffs on Chinese imports, Chinese state media outlets were told on May 6 to avoid mentioning U.S.-China trade tensions and posts about Trump’s tariff vow were swiftly deleted on WeChat and Weibo, two of China’s largest social-media platforms, according to The Wall Street Journal. Even privately owned media companies like Caixin made only veiled references to the tariffs.

Since Monday, local news media has ramped up their coverage of the trade spat, but articles have reflected Beijing’s defiant message that China won’t back down in a battle with the U.S.

See: ‘Bully us, wishful thinking!’ China preps for ‘people’s war’ as trade tensions mount

China’s broad-market equity index, the CSI 300 000300, -0.64% fell 5.8% on May 6, a day after Trump tweeted his tariff threat. That marked the benchmark’s biggest single-day slump since February 2016 when fears that a debt-heavy Chinese economy was on the brink of an abrupt slowdown took hold. China’s equity benchmark was down shy of 7% this month, trimming its year-to-date gains to 21%. The S&P 500 SPX, +0.80% is up more than 12% in 2019.

See: China’s financial market slump highlights sensitivity to Trump’s trade threats — in one chart

Analysts said government censorship presented a challenge for social media algorithms. Chinese internet users would employ memes and other roundabout language to circumvent restrictions on speech, making it difficult for text analysis to capture the various ways Chinese individual investors and research analysts would talk about trade jitters without drawing the state’s heavy hand.

For example, some Chinese social media bloggers referred to Trump as Thanos, the Marvel Universe supervillain, last Monday to allude to the U.S. president’s ability to devastate Chinese equities with a single tweet.

WeChat

Social media users describe Trump as Marvel supervillain Thanos

“Machines still often struggle with catching things like euphemism or sarcasm,” said Wool.

Still, Wool says recent advances in machine learning will eventually help algorithms to look through the more inventive ways humans use language to discover useful information, but that for now, such artificial-intelligence based techniques remained too slow to track how speech could mutate swiftly in online communities.

Providing critical information for the U.S. trading day. Subscribe to MarketWatch’s free Need to Know newsletter. Sign up here.

About the Author

Trade Stocks was specifically created as a resource that traders and investors can rely on for accurate and timely financial news and insights. Our team of industry experts brings you the into you need when you need it. That’s why hundreds of thousands of traders and investors trust us to provide them with the financial education, support, and resources that they can count on to help them move closer to their financial goals.