幾塊你要買？

1. P/E method — annualized EPS (Forward P/E)

The exchange's P/E is computed from trailing twelve-month EPS, which reflects the past year; for a company whose earnings are growing or shrinking, that lags reality. This method instead takes the year-to-date cumulative EPS from the latest quarterly filing and annualizes it:

annualized EPS = year-to-date EPS ÷ quarter number × 4
price = annualized EPS × your P/E multiple

Q1 ×4, Q2 ×2, Q3 ×4/3, Q4 is the full year. The fewer the quarters, the more extrapolation, so companies with pronounced seasonality or one-off items are easily distorted — the card says how many quarters were used. As with the trailing version below, the multiple defaults to this stock's own 5-year range.

Companies file on different schedules, so the quarter used differs by stock — the card label says annualized · 115Q2 or annualized · 115Q1 so you can see at a glance which filing the estimate rests on. Stocks that have filed the latest quarter use it; those that have not fall back to the previous quarter, so there is always a number.

Incidentally, the exchange publishes no P/E for companies that lost money over the trailing four quarters, so the trailing method below reports that it does not apply — but if the company has turned profitable this year, this annualized method still works.

2. P/E method — trailing twelve months (Trailing P/E)

price = TTM EPS × your P/E multiple

The most direct method; the multiple again defaults to this stock's own 5-year range. It does not apply to loss-making companies (negative EPS or no published P/E), and companies with volatile earnings are easily distorted.

3. PEG ratio

The blind spot of the P/E method is that it ignores growth entirely — at the same 20x P/E, a company growing 5% a year and one growing 30% are worth wildly different amounts. PEG puts growth in the denominator:

PEG   = P/E ÷ earnings growth rate (%)
price = estimated EPS this year × growth rate (%) × your target PEG

The literature often uses PEG = 1 as a reference point; the three target values are yours to set. Growth is taken as "estimated EPS this year ÷ last year's full-year EPS − 1", with a cap (40% by default) so that a very low prior-year base cannot produce an absurd target P/E. It does not apply when last year was a loss or this year is shrinking, and cyclicals are easily overvalued at the peak.

4. Monthly revenue momentum

Taiwan requires listed companies to publish the previous month's revenue by the 10th, making it the most timely fundamental data available — about 35 days ahead of the quarterly report, which is due 45 days after quarter-end. When earnings momentum turns, monthly revenue shows it first.

estimated annual revenue = year-to-date revenue ÷ months elapsed × 12
estimated annual EPS     = estimated annual revenue × (cumulative EPS ÷ cumulative revenue)
price                    = estimated annual EPS × your P/E multiple

The term in brackets is "how much EPS each dollar of revenue contributes", derived from the quarterly filing — so there is no need to look up the share count, and it automatically absorbs the average tax rate and non-operating items. It assumes the net margin stays stable, so companies whose gross margin swings will be off. Financial and insurance companies have no comparable revenue concept and are excluded.

5. Graham formula

The simplified formula from the revised edition of Graham's The Intelligent Investor:

V = EPS × (8.5 + 2g)

8.5 is the base P/E Graham observed for a no-growth company in 1962; g is the expected annual growth rate (%).

This site never presets g for any stock — you enter it under "Assumptions", and the card lists g ± your sensitivity setting so you can see directly how much the assumption moves the answer. EPS defaults to the estimate derived from quarterly filings and monthly revenue; you can override it with your own figure.

Worth noting: this is an empirical formula from the 1960s that ignores the interest-rate environment (the original also has a version adjusted by AAA corporate bond yields) and produces very high multiples for high-growth companies. Judge its applicability yourself.

5b. The Graham Number — not the same thing as the formula above

The difference first, because these two are easily confused. Method 5 above is the Graham formula; this is the Graham Number. Same author, opposite purpose:

In The Intelligent Investor Graham gives the defensive investor two thresholds: P/E no higher than 15 and P/B no higher than 1.5. Taken separately they leave a gap — a company at 8x earnings and 3x book passes one and fails the other, yet may be cheap on neither basis. So Graham combined them into a single constraint on the product:

P/E × P/B ≤ 15 × 1.5 = 22.5

and  P/E × P/B = (price ÷ EPS) × (price ÷ book value)
               = price² ÷ (EPS × book value)

so   price ≤ √(22.5 × EPS × book value)

That square root is the Graham Number. This site does not hard-code 22.5 — it pairs and multiplies the three P/E and P/B multiples you set under "Assumptions". Set those to 15 and 1.5 and you get Graham's original.

Who it treats unfairly: book value is in the formula, so brands, patents and software that never reach the balance sheet count for nothing. Asset-light software, IC design and distribution companies are flagged "too expensive" by construction — not because they are expensive, but because this formula cannot see their assets. It also assumes earnings and book value are stable, so cyclicals at a profit peak carry an inflated EPS and therefore an inflated ceiling.

10. Price / operating cash flow (P/OCF)

The nine formulas above all rest on earnings, book value, revenue and dividends — not one of them looks at cash. Earnings contain depreciation, amortisation and every kind of accrual, and can be adjusted without any cash moving. Operating cash flow is money actually received. A persistent gap between the two is usually worth investigating.

annualized operating cash flow = year-to-date ÷ quarters filed × 4
operating cash flow per share  = annualized figure ÷ shares outstanding
price                          = operating cash flow per share × your P/OCF multiple

TSMC is a good example: 28x earnings but only 21x operating cash flow, because its annual depreciation charge is enormous — an accounting expense that never leaves the bank. Capital-heavy industries generally look better on cash flow than on reported earnings.

Three limits to know up front.

First, this is operating cash flow, before capital expenditure. The US edition uses free cash flow (capex deducted); Taiwan cannot do the same, because the MOPS aggregate cash flow statement carries only the operating, investing and financing totals — no capex line and no depreciation, which would mean opening every company's filing notes one by one. So TSMC's 21x looks cheap, but its annual capex consumes most of that operating cash flow and far less is genuinely free. For capital-intensive industries you have to deduct that yourself.

Second, cash flow swings hard with working capital. Seasonal inventory and receivables move the figure sharply between periods — Hon Hai's cumulative operating cash flow for 115Q2 was actually negative, which does not mean it is unprofitable, only that stocking and payment timing fell that way. Annualizing from half a year is easily distorted; read it alongside a multi-year trend.

Third, it does not apply to financials. Bank cash flow statements are structured completely differently and large negative operating figures are normal, so the number cannot be read literally. About a third of the market has negative or missing operating cash flow; those report that the method does not apply.

6. Return on equity method

Uses how efficiently a company earns for its shareholders to infer what its book value should be worth. Derived from the Gordon growth model:

fair P/B = (ROE − g) ÷ (r − g)
price    = book value per share × the model's P/B

r is your required annual return and g your assumed long-run growth rate; both are your inputs. When ROE exceeds r the model returns a multiple above book value, and below r the reverse; the bounds then follow from your margin of safety. ROE must exceed g for this to mean anything.

7. Price / book (P/B Ratio)

price = book value per share × your P/B multiple

The multiple defaults to this stock's own 5-year P20 / P50 / P80. Suited to asset-heavy companies and those with volatile earnings (financials, construction, cyclicals), because book value is steadier than earnings.

8. Price / sales (P/S Ratio)

Revenue is always positive, so companies not yet profitable can still be valued — a group none of the earnings-based methods above can handle.

revenue per share = estimated annual revenue × (quarterly EPS ÷ quarterly net income)
price             = revenue per share × your P/S multiple

The term in brackets equals 1 ÷ share count (when loss-making, numerator and denominator share a sign, so it stays positive), which is why the share count never has to be looked up.

Excluded from the summary by default. Reasonable multiples vary enormously by industry — software and IC design often run 5–10x, distribution and contract manufacturing can be below 0.5x — and there is no universal default. Adjust the assumptions to the peer level first, then tick it in. It also ignores profitability entirely, so a company with high revenue and years of losses will be overvalued; always read it alongside the other methods.

9. Dividend method (discounted cash dividend)

price = dividend per share ÷ your target yield

Yield moves inversely to price, so the highest target yield gives the lowest price. The classic approach is 16x the dividend (6.25%), 20x (5%) and 32x (3.125%). Suited to mature companies with stable payouts; it does not apply to companies that pay nothing or pay erratically.

Adjusting for stock-dividend dilution

When a company pays a stock dividend the share count grows, so EPS and book value per share should be diluted in step. But on the ex-dividend date the exchange divides the already-diluted price by EPS that has not been restated for the new share count — testing Twinhead, Hwayu and Fartek, P/E and P/B all gapped 20–30% along with the price on the ex-date while EPS did not move at all.

Taken at face value, the EPS and book value this site derives would be too high, inflating the prices from the P/E, P/B and ROE methods alike. So whenever an ex-dividend date falls after the end of the reference fiscal quarter, this site divides EPS and book value per share by (1 + stock dividend ratio) to restore them, and flags it on the stock card. ROE is a ratio of the two and is unaffected.

Two known limits: for a combined stock-and-cash dividend on the same day, public data cannot separate the stock ratio from the cash portion, so this site flags it without correcting; and the historical valuation ranges carry the same stock-dividend discontinuities, so for frequent payers the "5-year range" runs slightly wider than reality.

Where the numbers come from

The TWSE and TPEx publish each stock's closing price, P/E, P/B and dividend yield daily, and this site derives the rest from those:

EPS                  = close ÷ P/E
book value per share = close ÷ P/B
ROE                  = EPS ÷ book value per share = P/B ÷ P/E
dividend per share   = close × dividend yield