[Originally published on October 28, 2020]

As the application of AI flourishes throughout various industries, what kind of new changes will it bring to the traditional investment industry?

Datayes! platform launched by Datayes has rolled out a set of smart investment & research frameworks to cover all listed A-share companies. With 1,000 in-depth VIP research frameworks, we are the sole player in China that has implemented AI-enabled earnings forecasts at scale.

Based on financial big data, Datayes! makes use of advantages of machine computational power to fully tap into the relationship between data and company fundamentals. By incorporating driving factors into the quantitative models and integrating the advantages of both human experts and machine intelligence, a set of mature man-machine integrated investment systems was created.

How to quickly perform earnings forecasts of 1,000 companies?

Rational investors normally adopt a set of systems to forecast an enterprise’s earnings, referencing high-frequency data regarding company operations, earnings previews, profitability estimates by research analysts and so on.

However, our data shows that there has been a substantial decline in the number of performance forecast announced by listed companies since 2019 after changes on performance forecast disclosure requirement were made by the Chinese stock exchanges.

Moreover, the number of stocks covered by research analysts also gradually decreased from 2017 as the attention of institutional investors became more focused.

Therefore, it is of great importance to forecast the performance of listed companies in a rational and forward-looking manner as publicly disclosed information is in decline.

Based on the theory mentioned above, Datayes has to date developed and published 1,000 in-depth VIP frameworks to forecast earnings of listed A-shares, through which their revenue can be monitored in real-time so that investment opportunities and early warnings are promptly issued. There are a total of 27 Shenwan Level-I sectors involved, including agriculture & farming, mining, chemistry, steel, nonferrous metals, machinery and equipment, national defense and military industry, automotive, electronics, etc.

In this way, Datayes! has successfully completed the “mission impossible” in the eyes of secondary market veterans, achieving the first and currently the sole scale application of AI-enabled earnings forecasting in China. It is also a dream come true to estimate profitability of over 1,000 companies during their non-disclosure periods.

On the interface of individual stock [AI earnings forecast], you can find the details of the forecast framework, analysis of the financial statements, absolute/relative valuation, factor sensitivity tests, etc.

Q1: How accurate is Datayes!’s AI forecasting?

Based on the research framework, our AI-enabled earnings forecasts were more accurate than the market consensus in previous trials.

• [2019 Annual Report] Success rate 82.5%：Datayes! bested the market consensus with its AI-driven earnings forecasting kit.
• [2020 Interim] Despite the impact of COVID-19 this year, the bias median of Datayes!’s AI-backed earnings forecasts was 2.07%. The bias narrowed below 1% in forecasting Robam and Wuliangye Yibin, showing decent accuracy. (Note: Interims and quarterly reports do not provide market consensus. Forecast statistics is for your reference only)
• [3Q20 forecasts ongoing] Take the 3Q20 estimates of Transfar Zhilian as an example. The latest announcement shows its revenue for the first three quarters declined 4.52% YoY to around RMB14.457bn. On October 16th, Datayes! published its earnings forecasts at RMB14.435bn, a decline of 4.67% YoY. The error stands at only 0.15%. (Note: 3Q20 estimates backtracking will be published in early November.)

Q2: What’s the difference between the newly-released VIP framework and the regular framework?

By introducing more human experts’ priori knowledge, the in-depth VIP frameworks have optimized research logics to better balance the model output accuracy and interpretability. Currently, there are around 1,000 VIP frameworks while the regular frameworks cover all sectors.

Datayes! Earnings Forecasting Walk-through

After top-down quantifying the “macro-industry-stock” factors impacting the fundamentals, Datayes! carries the relevant data into the framework, which will then forecast the earnings by leveraging advantages of AI computational power in data processing.