Within-Industry Relative Value (Market Neutral)

Thesis (edge)

Comparing an airline to a software company on the same valuation ratio is close to meaningless. They have different economics, different capital needs and different reasons for the numbers to look the way they do. A universal value screen ends up mostly betting on which industries are cheap, not on which companies are cheap.

The relative value approach narrows the comparison. Compare each airline only to other airlines. Compare each regional bank only to other regional banks. Inside a tight industry group, the companies face the same customers, the same input costs and the same regulatory environment. If one trades much more cheaply than its direct competitors without an obvious reason, that is a much more meaningful observation than any cross-industry ranking.

Then build the portfolio so the industry itself cancels: long the cheapest names in each industry, short the richest names in the same industry, in matched size. Whatever happens to airlines as a whole no longer matters. What matters is only whether the cheap airline outperforms the expensive one.

The edge comes from the tendency of valuation gaps between similar companies to narrow over time, and from stripping out all the noise you were never trying to bet on.

Where it works (regimes)

• Works well: in mature industries with many comparable companies, where the businesses really are similar and the valuation gap is more likely to reflect sentiment than structural difference.
• Works well: over months and quarters, giving the gap time to close. This is a patient strategy.
• Fails: when cheap is cheap for a reason. In a genuinely disrupted industry, the cheapest name is often the one being destroyed, and the expensive name is the one taking its business. Buying the former and shorting the latter is a reliable way to lose money for years. This is the single biggest weakness of the approach.
• Fails: in long stretches where valuation spreads simply widen. The expensive get more expensive and the cheap get cheaper, sometimes for years at a stretch. The strategy is right eventually and insolvent in the meantime, which is the oldest problem in value investing.
• Weakens: in industries with too few names, where the ranking is dominated by idiosyncratic noise.

Signals

• Valuation rank inside the industry. Use measures that actually fit the industry. Book value works for banks and is close to useless for asset-light software. Cash flow measures travel better than earnings measures. Use a small number of sensible ones rather than a large number of arbitrary ones.
• A quality filter. Pure cheapness ranking will hand you the most distressed company in every group. Adding a profitability or balance sheet strength measure filters out the names that are cheap because they are dying, which is the main improvement available here.
• The spread's own history. For a given industry, how wide is the gap between the cheap and expensive names compared to normal? Wide gaps offer more room to converge, though they can also signal that something structural is happening.
• Timing help. Some practitioners add a slow trend or momentum overlay so they do not buy a cheap name that is still actively falling. This directly addresses the "cheap and getting cheaper" problem, at the cost of missing some of the turn.

Portfolio construction

• Long and short inside each industry, matched by capital, so the industry exposure cancels.
• Neutralise the whole book against market beta and size at the aggregate level. Otherwise you can end up systematically long small distressed companies and short large healthy ones, which is a factor bet, not a relative value bet.
• Breadth: many industries, many names, small positions. Any single relative call is close to a coin flip.
• Rebalance slowly. Monthly or quarterly. Valuation gaps close over quarters, not days, and rapid rebalancing simply generates turnover without improving the signal.
• Cap turnover explicitly, and only trade when a name's rank has changed enough to justify the cost.

Risk model

• Value trap risk. The strategy systematically buys the cheapest companies. Some meaningful fraction of them are cheap because they are in permanent decline. No filter removes all of them.
• Spread widening risk. The gap you are betting on can widen substantially and stay wide. Drawdowns of this type are long and psychologically brutal, because nothing in the thesis is falsified while you lose money.
• Short squeeze risk. The expensive names you are shorting are often the market's favourites. Favourites can keep going up, violently, and the short leg can produce sharp losses that dwarf the slow gains on the long leg.
• Borrow cost. Persistent, and charged for the whole holding period. On a strategy with modest expected returns, borrow is a large fraction of the total.
• Classification risk. If your industry groups are wrong, the whole comparison is wrong. Companies get misclassified, and businesses change what they do without changing their industry code.
• Crowding. This is standard institutional territory. Positions are shared, and deleveraging is correlated.

Costs & implementation

• Turnover is moderate, which is one of the genuine advantages over faster mean-reversion strategies. Costs matter but do not dominate.
• Borrow is the main cost, not spread. Check it name by name before assuming a short is available.
• Data quality is the real work. Accounting data must be used with a proper reporting lag, restatements must be handled correctly, and the industry classification needs to be sensible. Most of the errors in this strategy live in the data pipeline rather than in the model.
• Look-ahead is easy to introduce. Using a company's annual figures from the day the fiscal year ended, rather than from the day they were published, is a subtle and very common way to manufacture a fake edge.
• Capacity is reasonably good compared with high-frequency mean reversion, because holding periods are long and the universe can include large names.

Failure modes

• Ranking on one ratio across all industries. That is not relative value, that is a sector bet in disguise.
• No quality filter. Buying the cheapest name in every industry means systematically owning distress.
• Ignoring the short leg's cost. Backtests that assume free shorting overstate this strategy substantially.
• Impatience. Cutting the strategy after a bad year is common, and bad years are structural here, not accidents.
• Overfitting the composite score. Blending eight valuation measures with weights chosen because they backtested well is a data mining exercise.
• Trusting industry codes blindly. A company classified as a retailer that is really a technology company will pollute both sides of the comparison.

Our Notes & Suggestions

Spend your effort on comparability, not on the scoring formula. The strategy works to the extent that the companies you are comparing are genuinely comparable. Time spent making sure a peer group is a real peer group is worth far more than time spent tuning the weights of a valuation composite.

Combine cheapness with some evidence that the business is not deteriorating. Valuation on its own selects for distress. Valuation plus a basic quality screen, or valuation with a check that the price is no longer falling, is meaningfully more robust and addresses the strategy's central weakness directly.

Be honest about the horizon. This is a strategy that can be right and still underwater for a long time. If your capital, or your patience, cannot survive a multi-year stretch of widening spreads, then the correct decision is not to run a diluted version, it is to accept that this is not the right strategy for you.