Value Investing Strategy: A Guide to Finding Undervalued Stocks

The margin of safety is the percentage discount between what you pay and what you believe the business is worth. The formula is arithmetic:

Margin of Safety (%) = (Intrinsic Value − Market Price) ÷ Intrinsic Value × 100

A worked example. A business earns $10.00 per share. Over the last decade the market has valued it between 15× and 25× earnings, averaging 20×. Nothing about the business has structurally changed, so 20 × $10.00 = $200 is a defensible estimate of fair value. The stock currently trades at $134.

Margin of safety = ($200 − $134) ÷ $200 = 33%.

That 33% is doing two separate jobs, and it is worth being explicit about both. First, it absorbs error: if your $200 estimate was optimistic and the business is really worth $150, you still paid $134 and did not overpay. Second, it is the return itself — if the market re-rates the stock back to $200, that is a 49% gain on the $134 you paid, before any earnings growth.

The margin of safety is always dependent on the price paid. It will be large at one price, small at some higher price, nonexistent at some still higher price.

Benjamin Graham, The Intelligent Investor

Note what the formula divides by. Dividing the gap by intrinsic value, not by price, is deliberate: it keeps the number anchored to your estimate of worth rather than to the market quote you are trying not to be influenced by.

Most practising value investors will not act below roughly 20–30%. But treating that as a universal threshold is the mistake. The right discount is a function of how confident you can be in the value estimate, and confidence varies enormously by business type.

The logic is that margin of safety compensates for estimation error, so it should scale with how much error is plausible. A 25% discount on a business you can forecast is worth more than a 50% discount on one you cannot.

One practical warning: if a screen is producing candidates at 60% discounts, the usual explanation is not that you have found something others missed. It is that the market is pricing in a deterioration you have not identified yet. Large discounts should increase your scepticism, not your position size.

Consider two companies, both trading at a P/E of 16. The first has traded between 22× and 38× for a decade. The second has traded between 9× and 18×. The same multiple means "unusually cheap" for one and "near the top of its range" for the other. The raw number carried no information at all; the position within the company's own distribution carried all of it.

This is why percentile context beats absolute thresholds. "P/E 16" is not a fact you can act on. "P/E 16, cheaper than 88% of readings over the past ten years" is.

Three things drive a company's normal multiple, and none of them are comparable across businesses:

• Growth rate — faster earnings growth mathematically justifies a higher multiple, which is the entire basis of the PEG ratio.
• Capital intensity — a business that must reinvest 80% of earnings to stand still is worth less per dollar of reported earnings than one that reinvests 20%.
• Earnings quality and predictability — a subscription business with 95% renewal rates deserves a higher multiple than a project-based business with the same reported earnings.

The practical rule: compare a company's multiple to its own history first, to close sector peers second, and to the broad market almost never. Cross-sector P/E comparisons are the single most common way beginners talk themselves into value traps.

A value trap is a stock that is cheap on every conventional measure and stays cheap, because the business is deteriorating fast enough to justify the price. The low multiple is not a mispricing — it is the market pricing in decline correctly, ahead of the accounts.

The distinction is not visible in the valuation metrics, because a trap and a bargain look identical there. It is visible in the business. Six red flags, in rough order of how reliable they are:

• Declining market share in an industry that is still growing. This is the single most damning signal — it removes every "it's just the cycle" explanation.
• Revenue per unit falling while volumes hold up. Pricing power is disappearing, and margin compression follows with a lag.
• Capital expenditure running below depreciation for several years. The company is harvesting the asset base rather than maintaining it, which flatters current cash flow and guarantees a future problem.
• A single product, patent or customer carrying the majority of profits, with a visible expiry date.
• The multiple has been low for years with no catalyst. If a stock has traded at 7× earnings for five years, the market is not making an error you just noticed.
• Management repeatedly missing its own guidance, or changing strategy every eighteen months.

Against that, what distinguishes a genuine bargain is a specific, identifiable, temporary reason for the discount — a litigation overhang that will resolve, a bad quarter in a cyclical business, a spin-off causing forced selling by index funds, or a sector-wide de-rating that caught a healthy company in the sweep.

The test that cuts through it: write down what has to happen for the gap between price and value to close, and roughly when. If the honest answer is "the market changes its mind," there is no catalyst and you are buying a trap. Cheapness alone has never been a catalyst.

For cyclical businesses — energy, mining, steel, autos, shipping, semiconductors, homebuilders — the P/E ratio does the opposite of what intuition expects. It is at its lowest when the stock is most dangerous, and at its highest when the stock is most attractive.

The mechanism is straightforward once stated. At the top of a cycle, earnings are at a peak that is not sustainable. Dividing a high price by peak earnings produces a low P/E. At the bottom, earnings have collapsed or gone negative, and dividing a low price by depressed earnings produces a high P/E — or none at all.

Peter Lynch put the rule bluntly: for a cyclical, a low P/E is bad news and a high P/E is often good news. The low trailing multiple is the trap signal, not the opportunity signal.

What to use instead

• Normalised earnings — average per-share earnings across a full cycle, typically 7–10 years, then apply a mid-cycle multiple to that average. This is Graham's own suggestion, adapted.
• Price to book, or price to replacement cost of the assets. Book value is far more stable than earnings through a cycle, which is exactly why Graham leaned on it.
• EV/EBITDA, which at least removes the distortion from leverage and depreciation policy, though it does not fix the cyclicality itself.
• Capacity and supply data. For commodity cyclicals, the years of underinvestment in new supply predict the next upturn better than any multiple does.

A discounted cash flow model is the theoretically correct way to value a business and, for most individual investors, the least reliable in practice. It multiplies three uncertain inputs — a growth path, a discount rate, and a terminal value — and produces an answer with false precision. Change the terminal growth rate by one percentage point and the output can move 30%.

Three alternatives give you a defensible number in a fraction of the time.

1. Historical multiple applied to normalised earnings

Take the company's median P/E over ten years. Apply it to the average of the last three years' per-share earnings. That is your fair value. It is crude, and it embeds the assumption that the business has not structurally changed — so check that assumption explicitly before trusting the output. For a stable business, it is usually within 20% of a carefully built DCF and takes two minutes.

2. The reverse DCF

Instead of forecasting cash flows to derive a price, take the current price and solve for the growth rate the market is implying. Then ask a single question you are actually qualified to answer: is that growth rate plausible? Discovering that a stock is priced for 14% annual growth for a decade is far more useful than producing your own point estimate of value, because it converts valuation into a judgement about the business rather than an exercise in spreadsheet arithmetic.

3. Earnings power value

Assume zero growth. Take normalised operating earnings after tax, divide by a required return of 8–10%, and add net cash or subtract net debt. This gives you the value of the business as it stands today, with no credit for future expansion. If the stock trades below that number, you are being paid to take the growth for free — which is as clean a value setup as exists.

The common thread: each method produces a range rather than a point, and each forces the assumptions into the open where you can argue with them.

If you do build a DCF, the discount rate is where most of the damage happens. Common practice lands between 8% and 15%, with the rate rising as business risk rises. The textbook answer is the weighted average cost of capital; the practical answer for an individual investor is the return you require to bother taking the risk, which for equities is rarely below 9%.

A workable default: 9% for a stable, low-debt, wide-moat business; 11% for an ordinary company; 13–15% for anything cyclical, leveraged or dependent on a single product. Do not fine-tune it to two decimal places — the input is not precise enough to justify that.

The four errors that ruin more DCFs than anything else:

1. 1Forgetting to discount the terminal value. The terminal value is a figure at year 10 and must be discounted back to today like every other cash flow. Adding it undiscounted is a mechanical error that can inflate the valuation by 60% or more, and it is the single most common mistake in beginner models.
2. 2A terminal growth rate above long-run GDP growth. Assuming a company grows at 5% forever means assuming it eventually becomes larger than the economy. Cap terminal growth at 2–3%.
3. 3Straight-lining current growth for ten years. High growth rates decay — competition arrives, markets saturate, the base gets larger. A ten-year forecast at a constant 20% is a statement that competition will never work, which the historical base rate firmly contradicts.
4. 4Tuning the discount rate until the model agrees with a conclusion you already reached. If you find yourself lowering the rate from 11% to 9% because 11% made the stock look expensive, the model has stopped being an analysis.

A useful discipline: build the model, then deliberately construct the bear case — growth 5 points lower, terminal growth 1 point lower, discount rate 2 points higher. If the stock is still cheap in that scenario, you have a genuine margin of safety. If it only works in the base case, you have a spreadsheet that agrees with you.

The Graham Number folds rules 6 and 7 of the defensive checklist — P/E ≤ 15 and P/B ≤ 1.5 — into a single ceiling price. Because 15 × 1.5 = 22.5, the formula is:

Graham Number = √(22.5 × EPS × Book Value Per Share)

Worked through: a company reports earnings of $4.50 per share and book value of $32.00 per share. 22.5 × 4.50 × 32.00 = 3,240. The square root is $56.92. Under Graham's defensive criteria, that is the most a conservative investor should pay. At $45 the stock offers a 21% margin of safety against that ceiling; at $70 it fails the test outright.

Two refinements make it substantially more useful. Use average earnings over the last three years rather than the trailing twelve months, which is what Graham specified and which removes most of the single-year noise. And check that book value is real — tangible assets and working capital, not goodwill accumulated from acquisitions.

Where it breaks

• It returns nothing usable when earnings or book value are negative — the product goes negative and the square root is undefined.
• It systematically rejects asset-light businesses. Software, pharma and consumer-brand companies expense the investments that create their value, so reported book value understates them, sometimes by an order of magnitude.
• It gives no credit for growth whatsoever. A business compounding earnings at 15% and one that is flat receive identical treatment.

Used as intended — a hard ceiling for defensive, asset-heavy, dividend-paying businesses — it remains a sharp tool. Used as a universal valuation model, it will steer you into exactly the capital-intensive, low-growth corners of the market where value traps concentrate.

Both cases break the standard toolkit, and both are more common than they were in Graham's era. They need different responses.

Negative earnings

A negative P/E is not a small P/E — it is an undefined one, and screens that rank by P/E ascending will happily sort loss-makers to the top of your "cheapest" list. Delete the metric and ask which of three situations you are in:

• Cyclical trough. Earnings are negative because the cycle is negative. Value on normalised earnings across the cycle, or on assets. This is frequently the best moment to buy.
• One-off charge. A write-down, litigation settlement or restructuring cost has pushed a profitable business into a reported loss. Add it back and value on underlying earnings — but verify the charge is genuinely non-recurring, because "one-off" charges that appear in four consecutive years are an operating expense with better marketing.
• Structural loss. The business does not make money and has no clear path to it. There is nothing to value conservatively here. This is outside value investing regardless of how far the price has fallen.

In the first two cases, EV/EBITDA and free cash flow yield usually still work, because a company can be unprofitable on a net income basis while generating real cash.

Negative book value

Negative shareholders' equity has two entirely different causes that look identical on the balance sheet. It can mean accumulated losses have exceeded paid-in capital — a genuine distress signal. Or it can mean the company has bought back so much stock, or paid out so much in dividends, that equity has gone negative while the business generates strong cash. Several large, healthy consumer and technology companies sit in the second category.

There is also a third, structural cause: accounting rules require research, development and brand-building to be expensed rather than capitalised. A company that has spent two decades building intangible value can carry that value on its books at zero. This is why P/B has steadily lost its power as a value screen — it now flags asset-light quality businesses as expensive growth stocks.

The resolution is to stop using book value for these companies and switch to cash-based measures: free cash flow yield, EV/EBIT, and return on invested capital. If P/B is the metric telling you a business is expensive, and the business has no significant physical assets, the metric is wrong, not the market.

Value strategies require patience, which creates an obvious problem: patience and stubbornness are indistinguishable from the inside, and the difference only becomes clear afterwards.

The honest answer is that the clock should run on the business, not the share price. Three to five years is a reasonable window for a re-rating, and a stock going nowhere for two years while the underlying business performs is normal, not a failure.

What should trigger an exit regardless of elapsed time:

• The falsification condition you wrote down at purchase has occurred. This is the whole point of writing it down.
• The value estimate itself has fallen. If you bought at a 35% discount to $200 and the business now supports $140, the discount is gone even though the price has not moved. This is the case people miss most often, because nothing dramatic happens on the screen.
• The catalyst you identified has resolved and the gap did not close. The market has now had the information and declined to act on it.
• Capital allocation has deteriorated — buybacks at high prices, an acquisition outside the core business, dividend funded by borrowing.

And the trigger that should not cause an exit: the price falling further with no change in the business. That is the situation the strategy exists to exploit. Distinguishing it from the second bullet above — where value has genuinely fallen — is most of the skill in value investing.