Updates & Corrections

When the home page was opened directly on a smartphone, by typing the address or reloading, the scroll-driven product videos did not load and stayed as black frames. They appeared only when arriving from another page on the site. The videos now load when the page is opened directly as well.

Where both a securities report and an amended report existed for one period, the later one was used. But some amended reports carry only the corrected passages and not the financial statements themselves. In production data, of 5,714 amended reports there were some containing not one balance sheet or income statement item. Simply taking the newer document let an empty one overwrite the original and left every figure blank. The document that has content is now preferred.

Some rows come from the provider with only a close, returning zero for the open, high, low and volume. Zero is treated as a valid figure, so the candle stretched from zero up to the close and the movement over that stretch was flattened out of sight. Dropping the rows entirely would leave months of blank chart, so they are now drawn as a point at the close with no range. We do not invent a range that does not exist.

Trailing-12-month figures already cover a full twelve months. On some paths, however, an annualization factor was applied to them again, so the trailing P/E, P/S and EV/EBITDA came out lower than they should for some stocks. Trailing-12-month values are now always treated as twelve months, and are no longer annualized a second time.

For some periods, gross profit was taken from a business-segment figure such as gross profit of other businesses, instead of the company-wide GrossProfit total. As a result, the gross margin came out lower than the operating margin, which cannot happen. This was most common among construction companies. We now use the company-wide total and have rebuilt the affected periods.

For banks and securities firms, deposits and borrowings are how the core business is funded, so net interest-bearing debt does not mean the same thing as it does for operating companies. Even so, we calculated and showed enterprise value, EV/EBITDA and net interest-bearing debt using the same formula as for operating companies. These sectors are now excluded from these metrics, and no value is shown.

For periods before a stock split or reverse split, EPS and book value per share stayed on the share count at the time of disclosure, while dividends are held on the current share count. The payout ratio and DOE calculated from the two were therefore off by the split ratio. We aligned both to the split-adjusted share count and recalculated all periods.

When we added a step that aligns the sign of earnings per share with the profit in the same results, it also treated items that are not profit amounts, such as equity-method investment gains and losses or the year-on-year change rate of net income, as profit. It could also compare figures from the preliminary earnings report (tanshin) against the annual securities report for the same period. As a result, EPS became negative for some profitable periods and the P/E ratio was not shown. We corrected what is compared and rebuilt the affected periods.

Some US stocks had days with no price, either because of the order in which new stocks were added and prices were loaded, or because a load temporarily failed. The next day's run fetched a different date, so those gaps were never filled. We reloaded the missing period and fixed the process so that a failed load no longer leaves a permanent gap.

The US stock treemap, screener, company comparison and stock search included preferred shares and foreign stocks traded over the counter (OTC) rather than listed on a US exchange. For preferred shares, the price belongs to the share itself while the financials belong to the issuer, so market cap, P/E and P/B became meaningless and could take up a large area of the treemap. We now limit these screens to stocks listed on NYSE, NASDAQ and AMEX and exclude stocks whose price and financials are on different bases.

Some companies prepare their financial statements in a foreign currency but disclose per-share figures in yen. For the latest quarter, book value per share was taken in the foreign currency and combined with the yen share price, so the P/B ratio was shown at a value far from reality. We now skip values whose currency does not match, and have fixed this.

Important announcements were pinned above newer ones regardless of date, so the list was not in date order, and the latest announcement shown on the home page was sometimes not the most recent one. Header images set on announcements were not displayed, and headings and lists in the body were not styled. All of these have been fixed.

Some calculation paths used net assets as-is for the equity behind ROE and related metrics, which included non-controlling interests and stock acquisition rights, so ROE came out lower the more subsidiaries a company had. The denominators also used only the period-end balance, which drifted from reality in periods with share issuance, buybacks or large acquisitions. We unified the definition of equity and now divide by the average of the opening and closing balances. We also stopped showing ROE for companies with negative equity, where dividing a loss by negative equity had produced a positive figure.

ROE, ROA and the equity ratio were still multiplied by 100 on the display side, a remnant of when the API returned them as a fraction between 0 and 1. The conversion stayed after the API had settled on percentages, so it was applied twice. Unit conversion now lives in the API alone and the display shows what it receives. Operating and net margins, which are calculated on screen from revenue, are unaffected because they are computed elsewhere.

When picking the latest full fiscal year, we did not check whether that period actually existed. For stocks that still had a row for a period that did not match their fiscal year end, that row could be chosen as the latest, giving revenue, total assets and share counts that were nothing like the real figures, so P/E, P/S, ROE and market capitalization were wrong in both the screener and the stock pages. We now prioritize periods whose length has been confirmed, and rebuilt the metrics for all stocks.

We kept loading the last price of U.S. stocks even after they were delisted, treating them as if trading were continuing. As a result, stale or incorrect prices were recorded for dates after delisting, and market capitalization and P/E could end up far from reality. Metrics combining a frozen price with newer financial results were also shown. We excluded delisted stocks from price loading and metric calculation, and removed the implausible prices recorded after delisting.

The opening balance on the cash flow statement is cash and cash equivalents, which differs from the balance sheet's cash and deposits: it includes short-term investments and excludes restricted deposits. When an opening balance was missing, the way we filled it did not check which statement a figure came from, so the balance sheet's cash and deposits could be used as the opening balance, and the opening balance, change and closing balance did not add up. We fixed this to use only figures from the cash flow statement.

A correction we had put in place to rescue misplaced digits was also being applied to correct values. For smaller companies with small figures, correct SG&A expenses and operating income were rewritten to one tenth of their value, and in some periods an operating loss was shown as an operating profit. We removed this correction and rebuilt the data using the values as disclosed.

A zero dividend is a real, disclosed state, but some of our processes treated it as 0% while others left it blank as unknown. As a result, for stocks paying no dividend, only the latest period could be blank, or the value could switch between 0% and blank depending on the time of day. We fixed this so that no dividend is shown as 0%, and a blank is used only when the dividend is undecided or unknown.

US results are often first published as preliminary figures on the announcement day and then corrected over the following days. Because we did not re-fetch a period once it had been loaded, pre-correction figures remained, and in some periods net income had the opposite sign from the actual result. Results that did not appear on the earnings calendar were also never loaded, leaving the latest period stale, and for companies whose income statement and balance sheet close a few days apart, ROE and ROA could not be calculated. We fixed this so that results are re-fetched shortly after announcement and missed results are detected from the data we already hold.

Several US stock metrics were each based on a different period or price. P/B was computed with the share price at each fiscal period end and did not follow the current price; the payout ratio divided year-to-date dividends by a single quarter's earnings; and P/S divided market capitalization by a single quarter's revenue. Book value per share was also divided by a share count that included potential shares not yet issued. We corrected these to use annual dividends, trailing-twelve-month earnings and revenue, the current share price, and the shares actually outstanding.

ADRs report their results in their home currency, while their shares trade in US dollars. We were dividing earnings per share in the home currency by a US-dollar share price, which produced P/E and related ratios far below their real levels. The same results can also arrive either in the home currency or converted to dollars depending on when they are retrieved, and we were summing quarters in different currencies into trailing-twelve-month figures. We fixed this so that ratios mixing currencies are no longer shown when the currencies do not match.

A browser draws its broken-image marker for an image that fails to load. On a server-rendered page the failure happens before any client-side handling begins, so it could not be suppressed. Images are now presented in a way that draws nothing on failure: the initial always sits underneath with the logo layered over it. The white ground beneath is there because some logos have transparent backgrounds, and without it the initial would show through.

The control to open it did not respond.

After a failure nothing was fetched again and the space stayed blank. It now retries.

With no way in, the chart could not be viewed at a larger size.

A column of figures alone made it hard to see whether a stock stood high or low against its peers.

The unit of the amounts was not stated, leaving no way to tell yen from thousands of yen.

As with P/E, PSR comes in a trailing and a forward form. No basis had been settled, so the trailing figure is now shown in preference.

US stocks held only strings formatted for display, and the CSV was written with notation such as "1.2T" as it stood. Japanese stocks wrote the underlying numbers, so the same action behaved differently depending on the market. The unformatted values are now carried through and US figures are written as numbers.

Periods with no line items were left out of the response, so the whole column for that year disappeared. On one stock, 2016 was followed by 2021. Missing years are now filled in as empty periods so the years run consecutively.

Digit grouping and the presentation of signs differed from the balance sheet and cash flow tables. All three now match.

After the API moved to returning percentages, a guess of "multiply by 100 if it is 1 or less" remained on the display side. That guess works backwards for a stock whose payout ratio exceeds 100%: a dividend of 100 yen against earnings per share of 10.79 yen is a payout ratio of 927%, and such stocks were shown at one hundredth. The guess has been removed and the value is shown as received.

Whether distributions are grouped by calendar year or by fiscal period depends on the security type, but that type was not passed from the display side and everything defaulted to an operating company. REITs and infrastructure corporations were therefore grouped by fiscal period, and the year shown could disagree with reality. The right grouping is now passed based on the security type.

The brokerages supported are all domestic, so the entry point is no longer shown where it does not apply.

Moving the shared component to trailing P/E alone left the P/E column blank on the US route, which does not return a trailing figure. The valuation element of the overview score also counts a missing P/E as zero, so instead of a blank it produced an incorrect reading of "expensive". Both were found and fixed before release.

P/E comes in a form based on trailing twelve-month results and one based on the company's forecast. These were handled as a single value, so which basis applied varied by screen. The screener, similar companies and the overview now each state which one they use.

The test for an article's source was loose enough to treat pieces distributed by outside media as our own, so articles from other news organisations carried our logo. The test now matches the one used on the US side.