Accounting numbers are prevalent in financial reporting, business valuation, and investment management. They’re so frequently used that the practitioners rarely asked pragmatic questions such as: are they useful, do they account for some meaningful risks, can they be used to price assets.
A recent article  attempts to bring some answers to these questions,
This paper is a treatise on handling accounting numbers in building factor models. Those numbers include book value, investment, return on equity (ROE), and other profitability measures that appear in standard models, but often from data mining without a clear explanation of why they indicate investment risk. The numbers are generated by accounting principles that deal with risk, providing an explanation but also a critique of how the numbers enter extant models.
The authors investigated the usefulness of book value, investment, and return on equity (ROE) numbers. They concluded,
…the expected return is given by the expected future earnings yield on current price, for which the current earnings-to-price, E/P, is a predictor. However, it is an imperfect predictor, for current earnings are typically disturbed by accounting principles that convey the risk to expected earnings that discounts their pricing. By recognizing this accounting and combining it with the current earnings-to-price, the paper produces a factor that explains stock returns well relative to existing models. That is complemented with an investment factor based on complementary accounting in the balance sheet under double entry. Accounting measures in standard models, like book value, investment, and ROE enter, but in a different way —in a way that conveys risk and return under accounting principles.
In brief, accounting numbers are useful, but only under some circumstances. This result is consistent with our experience in working with credit-risk models. We have observed that not all accounting numbers are useful in modeling probabilities of default. Their usefulness depends on the type of business and many other factors. Moreover, they must be used in conjunction with other types of market indicators such as macroeconomic, volatility, technical, etc.
The research results imply that we can incorporate accounting numbers into the development of investment strategies. But as in the case of credit-risk models, they must be used in conjunction with other types of market indicators.
Last and not least, by the same token, there is a growing number of people in the financial service industry who promote alternative data and many of them sell their data services at the same time. We ask ourselves again the same question: is the alternative data really useful. What do you think?
 S. Penman, XJ. Zhang, Accounting for Asset Pricing Factors, 2021, https://ssrn.com/abstract=3881177