Scoring without the metrics that could not be measured: 66 grades changed, so the rule was not adopted

Jupick uses nine metrics to grade a stock. When a value cannot be obtained, it does not enter 0 for that metric but inserts a value that scores exactly 50 points; why a neutral value is used rather than 0 is explained in another article on metrics that go blank by sector. This article takes up the next question: what happens if the metric is dropped instead of filled with a neutral value?

The case for dropping has a clear basis. A neutral value is ultimately a number Jupick made up, and when a made-up number enters the average, it dilutes the weight of the metrics that were actually measured. Scoring only on what is known, without filling in the unknown, looks more honest, and several data providers do exactly that.

The proposal was approved as far as its design in August 2026, with the final decision put off until one data cycle had run and the actual movement in grades could be measured. That measurement was run across all major stocks on the morning of September 1.

The results were not what was expected, so the proposal was not adopted. What was observed is recorded below, numbers as they are.

First, the design. The nine metrics are grouped into three categories: three for profitability, three for financial stability and three for dividends. Each category averages its own metrics, and the three scores are blended at 40% profitability, 35% financial stability and 25% dividends into a score out of 100.

When unmeasured metrics are dropped, a category average uses only the remaining metrics. That much is simple. The trouble comes next. If a category has too few measured metrics, its score cannot be trusted, so the whole category is dropped when its measured count falls below a set number. The design under review set that threshold at 2, meaning at least 2 of the 3 metrics must be measured for the category to count.

When a category drops out, the weight it held has nowhere to go. The remaining categories refill 100% while keeping their ratio. If financial stability drops out, profitability rises from 40% to 61.5% and dividends from 25% to 38.5%.

This redistribution is the whole story here. If profitability and financial stability drop out together, the dividend category alone holds 100%. That is, three dividend metrics alone decide the company's grade.

Jupick read the major stocks stored as of the morning of September 1 as they were, scored them under both the current rule and the new rule, and compared the results. The grade letter changed for 66 stocks (16.5%). Of these, 54 went up, 5 went down and 7 lost their grade entirely.

Up moves outnumbering down moves about ten to one was the first warning sign. A rule meant to handle missing values honestly has no reason to push results in only one direction.

Splitting the 66 cases again, 35 came with a category dropping out and 31 changed without one. The 15 largest gains all, without exception, came with a category dropping out. By contrast, the 5 that went down all had no dropout and only crossed a grade boundary by 1 to 5 points.

Jupick also counted how many dropouts occurred across all major stocks, not just those whose grade changed. Financial stability dropped out most often, 37 times, followed by dividends 27 times and profitability 18 times. A dropout often happens without changing the grade letter, so this total exceeds 66. It means that many stocks cannot get 2 of the 3 financial stability metrics measured.

One pattern repeats in the list of largest gains: stocks with 1 profitability metric, 1 financial stability metric and 3 dividend metrics measured. With a threshold of 2, profitability and financial stability drop out together and only dividends remain.

Stocks with this pattern took 10 of the 12 largest gains. Samsung Securities rose 35 points, from B at 59 to S at 94; Korean Reinsurance and Industrial Bank of Korea rose 34 points each, Samsung Life Insurance 32, and Samsung Card and iM Financial Group 31 each. DB Insurance followed with 28 and KB Financial Group with 27.

The grades given here are not grades Jupick assigns. They are the output of an experimental calculation that was not adopted; they have never been shown on screen and never will be. They are cited to show how much a single rule moves the results.

Securities firms, insurers and banks clustering on the list is no coincidence. Financial stability metrics are hard to capture for the financial sector in the first place, and the reasons are set out in the article on metrics that go blank by sector. To carry over only the result: these stocks yield just 1 measured metric in the financial stability category.

The problem is that the new rule treats this not as “financial stability is not assessed” but as “assessed on dividends alone.” A financial company that pays steady dividends earns the top grade because its financial items cannot be measured. The fact that something cannot be measured gets translated into good news.

Judging a threshold of 2 too aggressive, Jupick reran the same calculation with a threshold of 1, a setting that keeps a category as long as it has at least one measured metric.

This time 58 stocks changed grade: 37 went up, 16 went down and 5 lost their grade. Only 9 came with a category dropping out, and the largest gain shrank to 14 points. The biggest gainers were JPMorgan Chase and PNC Financial Services, both up 14 points from A to S.

The problem was the list of 16 that went down. The grades here, too, are not grades Jupick assigns but the output of the experimental calculation that was not adopted. Samsung Life Insurance fell 14 points from B at 51 to C at 37; Korean Reinsurance and KB Financial Group fell to C at 49 each, Industrial Bank of Korea to C at 46, and Samsung Card and iM Financial Group to C at 45.

The meaning shows only when this is read alongside the previous section. On the same day with the same data, Samsung Life Insurance gains 32 points with a threshold of 2 and loses 14 with a threshold of 1. Korean Reinsurance, Industrial Bank of Korea, Samsung Card, iM Financial Group and KB Financial Group all go through the same thing. Changing one number in the rule flips the verdict on at least these six stocks entirely.

The flip comes from the financial stability category. With a threshold of 1, the category survives, but its surviving score is built from a single measured metric. What used to be softened by averaging with two neutral values now makes up the entire category score on its own. All six stocks are financials, so the one remaining metric is likely the debt ratio, which in the financial sector does not read the same way as in manufacturing and tends to score low. This measurement, however, counted only the number of measured metrics per category and did not check which metric remained. So far this is an inference.

Setting the two configurations side by side shows where the problem lies. The arithmetic of dropping missing metrics from an average is mild in itself. Cases that changed without a category dropping out only edged back and forth across grade boundaries.

What upended things was dropping whole categories and redistributing the leftover weight. That is where a 25% category suddenly becomes 100%, and 1 metric out of 3 stands for the whole category. A rule meant to handle missing data honestly becomes one that amplifies the remaining metrics to extremes, and the more data a stock is missing, the stronger the effect.

More important, the direction is not settled. With a threshold of 2, financial stocks rise to the top; with a threshold of 1, the same stocks drop toward the bottom. Both settings have their own logic, and the data does not say which is right.

When verdicts are this sensitive to the choice of rule, it signals not a poorly chosen rule but a lack of material. A framework that measures financial stability with three metrics does not fit banks and insurers, and whatever missing-data treatment is laid on top of it, the treatment ends up deciding the result.

The experiment produced one unintended result. With a threshold of 2, 7 stocks lost their grade entirely, and with a threshold of 1, 5 still lost it.

Failing a threshold of 1 means no measured metric in any of the three categories. In other words, all nine metrics are fill-in values.

Yet these 5 stocks were showing C 46 on screen as of September 1. Not a single piece of actually measured financial information on them went into the grade calculation, and still a grade letter went out with them.

Jupick also traced where 46 came from. A fill-in value is chosen to score exactly 50 points, so if every value is a fill-in, the score should be 50. But of the nine metrics, the payout ratio alone could not get such a value. The payout ratio loses points whether it is too high or too low, so its scoring curve is shaped like a hill that peaks in the middle, and the 50-point mark appears twice, once on each side of the hill. 15% scores 50 points and so does 55%. Choosing either side would make a different claim, so neither was chosen and the metric was left at 0.

On this curve, 0% scores 12.5 points. The dividend category becomes 37.5 points, the average of two 50s and one 12.5, and weighting profitability at 50 by 40%, financial stability at 50 by 35% and dividends at 37.5 by 25% gives 46.875 points. That is what 46 was. A stock with no information was starting not from neutral but from a little below it.

This was fixed on September 3. Instead of choosing one of the two candidates by preference, Jupick looked at the distribution of the 324 stored major stocks (81%) that actually had a payout ratio. The median was 31.19% and 55.2% sat below the ideal of 35%, so 15% is closer than 55% to the group Jupick covers. The anchor value was therefore set at 15%, and a stock whose nine metrics are all fill-in values now totals 50.0 points, not 46.875.

Whether a grade letter should go out at all for a stock with zero measured values, however, remains a separate question. This round got as far as confirming and recording the facts.

The conclusion is not to adopt it. The grade calculation rule stays as it is.

There are three reasons. First, most of the change the new rule produces comes not from better data but from redistributed weight. Second, where to set the threshold is an arbitrary choice, and that choice flips the verdict on at least the six stocks confirmed so far. Third, the stocks that move most under the new rule are concentrated in exactly the sectors Jupick has said all along it cannot measure well.

Changing the rule does not make things better; the lack of data just shows up in a different shape. When a rule is used to fill in missing material, the result reflects the rule rather than the material.

The current method is not kept because it is accurate. Filling with neutral values is imprecise too. But that imprecision has no direction and, above all, is not hidden on screen. Where a fill-in value is used, a hyphen appears in place of a number and the reason for the blank is shown beside it. The alternative under review gave the imprecision a direction while hiding that fact behind the grade letter.

Jupick applied the proposal to drop unmeasured metrics and score only what remains to its major stocks on September 1, 2026.

With a category threshold of 2 measured metrics, 66 stocks changed grade and 54 of them went up. The largest gains went to financial stocks where profitability and financial stability dropped out together and dividends got 100% of the weight, with gains of up to 35 points.

With the same data and only the threshold lowered to 1, 58 stocks changed and 16 of them went down, and the financial stocks that had gained the most before were now on the list of those that fell.

What drove the change was not the arithmetic of dropping missing metrics but dropping a category and redistributing the leftover weight. If a single threshold value flips the verdict on the same stocks, the problem lies not in the rule but in the material.

So the proposal was not adopted, and the current neutral-value treatment and hyphen display remain. One separate fact was confirmed: 5 stocks have all nine metrics as fill-in values, and on screen they carry a grade that looks the same as any other stock's.

This content is for information only and is not investment advice. Jupick gives no trading advice and no price forecasts, and the stocks and grades used as examples serve only to explain the properties of a calculation method that was not adopted; they are not an assessment or advice of any kind. The experimental grades in this article are not grades Jupick provides. The figures cited were measured on September 1, 2026 and change with updates. Investment decisions and responsibility for them rest with the user.