Russia’s retailed equity boom is gathering momentum and it is, like elsewhere, increasingly being driven by social media, according to a survey carried out by
After six years of almost continuous rate cuts by the
The CBR has long complained about the low level of financial literary of the population, but thanks to the deep penetration of the internet – there are more people online in
The coronacrisis has only accelerated the process as Russians forced to stay at home have spent more time playing with their phones and computers when looking for somewhere to park their savings and protect them from the ravages of high inflation and the instability of the ruble.
As recently featured by bne IntelliNews, the
“Russian retail stock market investors have become a mature and material investor base whose investment patterns and preferences have considerable influence on market outcomes,”
“We attempt to gain insight into the sentiment of the collective retail investor base at a granular level by studying unstructured text data from
Retail investors now account for an estimated 40% of the turnover on the Russian stock market and are increasingly affecting its performance. For example, in September international investors sold down some of this year’s gains, but Russian retail investors were still buying aggressively, keeping the market buoyant, according to BCS GM. International investors selling should have reduced the exposure to
VTBC's findings show that for a subset of publicly traded stocks, data suggest a potential causal link between social media sentiment and price changes, and so insights into social sentiment may help inform professional investors’ asset allocation decisions.
Below we reproduce the set of questions and answers from the VTBC survey that were released in a report on the impact of social media has on Russian retail investment decisions:
Q: How large is the retail investor base in
Q: You want capture retail sentiment – but in what sense?
A: Yes, we attempt to capture social sentiment toward individual stocks.
More practically, we harvest publicly available text data at scale, identify discussions about publicly traded companies, and map these onto a five-point scale ranging from 'strong sell' equivalent to 'strong buy' equivalent messages, and everything in between.
Q: How?
A: In short, public social media messages + modern text-processing machine-learning methods.
More particularly, we track text updates posted by financial market-focused groups with meaningful audiences (i.e. 1k+) in one of the most popular social media platforms in
We then fit a set of text classification ML models to a manually scored seed message sample, with our models ranging in complexity from standard (i.e. regularised logit) to state-of-the-art (i.e. a classification neural net head on top of the deep learning Russian language model SBERT). The resulting sentiment score is an ensemble, i.e. a consensus, of our models.
Q: What is the scale of the exercise? Is it representative?
A: The total audience count covered is some 1.5mn, which is approximately 11% of the unique and 75% of the active retail investor accounts on the
The audience of the channels (channels are essentially author-centred communities where messages are translated onto the audience) definitely has a substantial intersection and the unique audience is substantially smaller. Still, we believe that the universe of opinion is adequately covered.
We deliberately concentrate on a segment of social media that is focused on stock picking and are not general 'news aggregators' in order to reduce signal-to-noise ratios.
Nevertheless, even in the stock picking community, 'statement of fact without price crossread' messages dominate the flow.
Q: What is the key takeaway from an equity investor perspective?
A:This depends on the investor – and the covered sector – as stock prices, on our numbers, have varying degrees of sensitivity to public sentiment.
In order to estimate this sensitivity, we run a Granger causality test for 15 lags of social sentiment influencing stock price changes and calculate the corresponding pvalues. The null hypothesis here is that there is no causality, so the lower a stock’s pvalue, the stronger the evidence that the influence of public sentiment on that stock's prices should not be excluded.
The bar chart below left shows some of the stocks that have the lowest mean pvalues in our sample covering
Q: What is current social sentiment by stock?
A: In the chart below we exclude neutral sentiment (i.e. messages concerning statement of fact) and only show those with strong current sentiment. We also censor tickers with the most negative social sentiment.
Q: Did you test how a hypothetical portfolio allocation strategy based on social sentiment would perform?
A: Yes. The charts below show a stock sentiment-based single stock strategy performance relative to 'buy and hold' for January 2019 -
For illustrative purposes we estimated a strategy in which short stocks are 'disliked' or 'hated' and longs stocks 'liked' and 'loved'. The charts below show that for some of the stocks, social sentiment outperforms simple buy and hold and for others trails behind it.
©2021 bne IntelliNews
, source