See Wahlen, J. and M. Wieland. 2011. Can Financial Statement Analysis Beat Consensus Analysts’ Recommendations? Review of Accounting Studies 16, 89-115.

Source: Thesis by Jing Yuan for the Masters Program in Accounting and Fundamental Analysis, Columbia Business School, Effectiveness of Security Analysts’ Recommendations: Insights from China and the U.S., December 2024.

Distribution of Ratings

The following graphs summarizes the distribution of ratings. The ratings use the I/B/E/S standardized IRECCD ratings ranging from 1 to 5, where 1 represents Strong Buy, 2 represents Buy, 3 represents Hold, 4 represents Sell, and 5 represents Strong Sell. CSMAR applies standardized 1-5 recommendation ratings for Chinese analysts: 1 represents Buy (买入), 2 represents Overweight (增持), 3 represents Hold (持有), 4 represents Underweight (减持), and 5 represents Sell (卖出). The 1-5 rankings are broadly comparable.

Ratings in both countries are shewed towards BUY, more so in China.

Sell recommendations are relatively rare. Is this because the firms that analysts select to cover are those with better prospects? Is it because analysts tend to have in upward bias? Is it because analysts associated with brokerage houses are inducing us to trade?

Recommendation revisions are viewed as the most informative feature of analysts’ reports. The following gives the frequency of recommendation upgrades and downgrades from one level to another.

The U.S:

China:

Do analysts’ recommendations revisions predict stock returns? Here are the findings from this sample period. The tables give the average return for holding periods following the revision. Returns are excess (abnormal) returns over the market return, that is, the return for the target stock minus the return on the market for the period. *** indicates statistical significance at the 1% level, ** at the 5% level.

U.S.:

China:

A 2024 paper asks whether a LLM can do financial statement analysis and, from that analysis, develop credible earnings forecasts. It feeds financial statements (without companies’ names) to GPT4 and finds the artificial intelligence predicts earnings with more accuracy than analysts. Analysts are a good benchmark for they are not only human intelligence but supposedly have expert systems. Of course, it is well known that analysts also have biases and institutional incentives that affect their forecasts.

See Kim, A., M. Muhn, and V. Nilolaev. 2024. Financial Statement Analysis with Large Language Models. Chicago Booth Research Paper. At https://ssrn.com/abstract=4835311.

Of course, earnings prediction is only part of the task. One has to convert the forecast into a trading strategy.

In valuation, the investor asks: What will future financial statements look like? The investor then uses those predictions to challenge the market price: Can the market price, particularly the pricing of growth expectations, be justified by pro forma future financial statements? The following paper applies AI to the task:

Geertsema. P., H. Lu, and G. Ma. 2024. Projecting Financial Statements with Artificial Intelligence. At https://ssrn.com/abstract=5039433.

Forecasting score = w₁A₁ + w₂A₂ + w₃A₃ + ……

where the A numbers are the drivers and the w are the weights given to the drivers.

This has been attempted, though can probably be improved upon. (Get to work!) The score is called an S-score (an income sustainability score) with a high S-score indicating a higher probability of RNOA increasing and a low S-score indicating a higher probability of RNOA decreasing.

Here is how the S-score tracks RNOA:

Year 0 is the year that the financial statement analysis to calculate the S-score is carried out. A high S-score is one in the top 1/3 of scores, a low S-score is one in the lowest 1/3. In the plot, the S-scores forecasts the actual RNOA five years ahead. The difference between the future RNOAs for low and high S-scores is quite significant, about 4%.

The S-score also predicts stock returns. Here are the average difference between returns from investing in high S-scores versus low S-scores, adjusted for firm size:

These are returns from going long on high S-scores and short on low S-scores. They are positive almost every year. The graphs need to be updated for post-2002 years.

See Penman, S., and X. Zhang. 2006. Modeling Sustainable Earnings and P/E ratios with financial Statement Information. At https://ssrn.com/abstract=318967.

Prospective financial statements give the analyst a better handle on the forecasting. They are simply (!) the full financial statements estimated for the future, displaying the forecasted evolution of the statements from the current financial statements. Subject to the same chapter 7 financial statement analysis of current and past financial statements, they yield a more detailed analysis of drivers. So, for example, the future operating liability leverage, the gross profit margin, the inventory turnover, the R&D-to-Sales ratio are explicitly modeled. The prospective financial statement modeling also reveals to the investor his uncertainty: What prospective driver do I have difficulty in getting a hold on? That bears on his assessment of risk in chapter 8.

The investor maintains these prospective financial statements, updating them as more information arrives and consequently her buy or sell decision. One such piece of information is new financial statements published. The investor not only revises the prospective financial statements based on actual realization of the line items but also observes the variance of actual versus estimate: Where did I go wrong? Learning over time.

The formula also gives us a short cut for forecasting future residual income. If our financial statement analysis indicates that particular drivers are fairly constant, then forecasting is simplified:

• If ATO is constant, future ReOI is given by a forecast of Sales and PM
• If PM is constant, future ReOI is given by a forecast of Sales and ATO
• If PM and ATO are constant (and thus RNOA is constant), future ReOI is given by a forecast of Sales

S&P Global has developed a Moat Index:

https://www.spglobal.com/spdji/en/indices/dividends-factors/sp-500-economic-moat-index/?utm_source=pdf_research#overview

This index quantifies the moat abstraction by rating firms on three measures:

• Sustained ROIC (similar to RNOA): The average over the past five years
• Sustained gross margins: Gross margin/Revenue and its standard deviation over the past five years
• High market share: Weighted-average revenue in industry, calculated with Syntax’s Market Share Score

Your authors have not validated the effectiveness of the score but S&P report slightly higher returns to identified moat firms over the S&P 500.

See the S&P Moat Index fact sheet at file:///C:/Users/shp38/Downloads/fs-sp-500-economic-moat-index-usd.pdf

Another attempt to measure the moat is in Mauboussin, M. and D. Callahan, 2013. Measuring the Moat at http://csinvesting.org/2013/07/24/measuring-the-moat/

Data analytics and information aggregators are firms often hailed as castles with moats around them. S&P Global (or its parent company, Moody’s Corporation) is one, as are FactSet Research Systems, MSCI, Equifax, TransUnion, and Fair Isaac. There firms have proprietary data, proprietary methods of analyzing data, and supply benchmark metrics that many employ. They have sustainable revenue with subscriptions and high margins.

These firms trade at high valuations. Fair Isaac, the provider of the FICO score, traded at a forward P/E of 78 at the end of 2024, for example. That might be because of potential growth. But the value investor is cautious: Those multiples could be overpricing. Is the expected growth justified? Is it at risk? Should the P/E be discounted for that risk? Is the advent of AI to elicit and analyze information a breach of the moat?

• High ATO comes with low operating PM (Restaurants, Business services, Trucking, Food stores, General stores, Wholesale durable goods)
• Low ATO comes with high operating PM (Pipelines, Communications, Utilities, Oil and Gas, Railroads)
• High financing leverage (FLEV) comes with low operating liability leverage (OLLEV) (Pipelines, Communications, Utilities, Hotels)
• Lower levered profitability (ROCE) is largely due to lower RNOA, not lower leverage.

These relationships are just averages, not necessary for individual firms.

This complements the ATO-PM plots in chapter 7. The graph below is similar to the one in chapter 7 but with different industries indicated:

• A firm can have a high P/B but zero growth (Cell D in the tic-tac-toe game)
• A firm can have a high P/B but negative growth (Cell G in the tic-tac-toe game)
• A firm can have a low P/B but positive growth (Cell C in the tic-tac-toe game)

An example of the first two cases is a pharmaceutical firm with high P/B because its R&D asset is missing from the balance sheet. It reports positive residual earnings (RE) it is not expected to grow or even decline. The decline could be due to the R&D no longer delivering residual earnings as high as currently…lower profitability from the R&D pipeline. This can be seen from the valuation formula for enterprise value:

Price can be considerably higher than NOA₀ on the balance sheet with positive (RNOA₁ – r).NOA₀ but g can be zero or negative.

The 3 x 3 is a combination of P/E and P/B, each broken up into three levels, high, normal, and low. A normal P/E is one with no growth, that is, P/E =  which is equal to 10 if the required return is 10%. A normal P/E is also one with no growth in residual earnings. So high P/E above normal (for example) is one where residual earnings are expected to grow. A normal P/B = 1.0, that is, a P/B where expected future residual earnings are zero. So a high P/B (for example) is one with positive expected future residual earnings.

So, fill out the cells!

Here is the solution. (Don’t look before you have played the game!)

We played this game some time ago. Here are companies slotted into the respective cells by comparing the expected future RE to the current RE:

What do we learn from this?

• • P/E indicates growth in RE but this could be from a very low base: Firms in cell C with low P/B can be high P/E firms
  • Trailing P/E reflects growth and transitory earnings. If earnings are temporarily low, P/E will be high
  • The Molodovsky Effect:
    • Cells B and H are pure Molodovsky effects: current earnings include one-time unsustainable transitory income which cannot grow. So a high P/E could be due to a one-time impairment charge thst reduces earnings, for example.
    • Cells A, C, G and I are mixed growth and Molodovsky effects