Glossary
The terms, in plain English
Backtest #
A strategy's simulated performance on data from a past it was designed while looking at — a hypothesis wearing the costume of a result.
A backtest is the simulated performance of a strategy applied to historical data — data that existed, and was usually studied, before the strategy was finalised. That ordering is the whole problem. With enough adjustable rules, any past can be fitted perfectly, including its random noise; the more a backtest was optimised, the less it predicts, because the fit captures coincidences that will not repeat. Serious evaluation demands out-of-sample and walk-forward evidence: rules frozen first, then run on data they never saw, across more than one regime, with enough independent trades to separate skill from a coin that landed heads. The trap is typographical: “the strategy returns 22 percent a year” in the headline, “backtested” in the footnote or absent. Two questions separate evidence from theatre — is this record live audited money or a simulation, and does it span a genuine crash?
Basis point #
One hundredth of one percentage point — the unit finance uses when small differences compound into large ones.
A basis point is one hundredth of one percentage point: 50 basis points is 0.50 percent; 200 basis points is two percent. The unit exists because in markets, differences that sound trivial compound into fortunes. An annual fee difference of around 180 basis points, applied to a portfolio compounding at seven percent for twenty-five years, consumes roughly a third of the total gain — without a single bad market year, because fees are the one certain number in a sea of uncertain ones. Thirty basis points of credit-spread widening can mark the early stage of a regime change. The trap is framing: costs are quoted in basis points precisely because “ninety basis points” registers as smaller than “nearly one percent of the money, every year, regardless of performance.” Whenever a cost appears in basis points, restate it as a yearly amount on the actual sum involved.
Correlation #
A measure of how much two assets move together — the number diversification depends on, and the one that breaks precisely in crises.
Correlation measures how much two assets move together, from minus one (opposite) to plus one (lockstep). Diversification is built on low correlation — and the number is regime-dependent in the worst possible way. Measured on calm days, many assets look unrelated; measured during crashes, correlations converge toward one (Longin–Solnik), because the same leveraged holders face margin calls everywhere at once and sell whatever they can. Assets fall together not because they are linked, but because the forced sellers are the same people. In 2022 the stock-bond correlation flipped positive and the classic “balanced” portfolio lost on both sides simultaneously. The trap: portfolios are marketed as “diversified” using correlations computed across calm periods. The honest question is conditional — what is the correlation of these holdings given a twenty-percent market fall? A diversification claim that cannot answer it has only described fair weather.
Covered call #
Selling away the right to an asset's big gains in exchange for a steady premium — 'income' that is really a capped upside plus a kept downside.
A covered call is the sale of a call option against an asset already owned: the seller collects a premium today and gives away the asset’s gains above the strike price. The premium is real cash, which is why covered-call funds advertise high “distribution yields.” But the structure is a short volatility position in disguise: the holder keeps the full downside, caps the upside, and so underperforms exactly in the strong rallies that generate most long-run equity returns. The higher the advertised yield, the more upside has been sold to fund it — yield and expected total return trade off mechanically. The trap is the word income. A distribution funded by selling upside is not return; the payout and the forgone gains net against each other over full cycles, and marketing quotes only the first half. The completing question: what total return did this deliver against simply holding the asset?
Credit spread (OAS) #
The extra yield a risky borrower pays over a government bond — the market's live price on the chance of not being paid back.
A credit spread is the extra yield a risky borrower must pay over a government bond of the same maturity; the option-adjusted spread (OAS) is the refined version that strips out embedded options. The spread is the market’s continuously updated price on default risk — and one of the earliest honest warnings available. Investment-grade spreads widening by thirty basis points from their low, with no obvious news, has historically flagged trouble months before equity indices react: credit quietly cracking while headlines stay calm (the Adrian–Boyarchenko vulnerable-growth work formalises this). The trap: high-yield products are marketed on the yield and stay silent on the spread, when the yield is the default risk being paid for. An eight-percent “income” bond is not a discovery; it is a borrower the market charges eight percent.
CVaR (Conditional Value at Risk) #
The average loss across the worst few percent of outcomes — not where the bad region begins, but how deep it runs.
Conditional Value at Risk (CVaR, also called expected shortfall) answers the question VaR refuses: given that one of the worst five percent of outcomes has arrived, what is the average loss across those scenarios? It measures the depth of the bad region rather than its doorway, and unlike VaR it aggregates honestly — it satisfies the coherence axioms (Artzner; estimation per Acerbi–Tasche), so diversification can never appear to increase risk. The practical question it enables: in the worst months, what is the average loss, and is that survivable without selling at the bottom? The trap: a CVaR is only as deep as the worst event in its sample. Computed on a short, calm history it looks reassuring, and it is blind to losses that never appear in price data — an issuer default, a frozen fund, a custody failure. A risk number cannot see what was never in the sample.
Drawdown #
The deepest fall from a peak to a trough an investment has suffered, and how long it took to climb back.
A drawdown is the fall from an investment’s peak value to its subsequent trough, usually paired with the time it took to recover. It is the pain number: a strategy averaging ten percent a year with a fifty-percent worst drawdown demands that the holder watch half their capital vanish and keep holding anyway. The arithmetic is unforgiving — a fifty-percent loss requires a one-hundred-percent gain just to break even, because the climb starts from a smaller base. An index that falls 34 percent in five weeks, as global equities did in early 2020, tests the holder more than any average return reveals. The trap: pitches lead with gains and omit the drawdown entirely, or show a record too short to contain a real crash. A performance chart with no drawdown figure is a swimming pool advertised without depth markings.
Duration #
A measure of how much an asset's price falls when interest rates rise — the longer the wait for its cash flows, the harder rates hit it.
Duration measures how sensitive an asset’s price is to a change in interest rates: roughly, a duration of seven means a one-percentage-point rate rise cuts the price by about seven percent. It applies well beyond bonds — a growth stock is a duration asset in disguise, because its value sits in distant cash flows that higher rates discount away. That is why 2022 hurt “balanced” portfolios so badly: equities and long bonds were the same rate bet wearing different labels, and both fell together. The trap: bond funds are routinely sold as the “safe” portion of a portfolio without stating their duration, and a long-duration fund can lose double digits in a rate shock. Safety language without a duration number describes calm weather, not the vehicle.
Kelly criterion #
A formula for how much of a bankroll to stake on a favourable bet so that long-run compound growth is maximised — and the reason professionals deliberately bet less than it says.
The Kelly criterion (Kelly, 1956) computes the fraction of capital to stake on a favourable bet so that long-run compound growth is maximised: more when the edge is larger, less when the downside is wider. Its deeper lesson is that over-betting is fatal even with a genuine edge — stake beyond the Kelly fraction and compound growth falls, eventually below zero, because losses compound from a shrinking base. Since the inputs (the edge, the odds) are always estimated with error, practitioners deliberately size at half or quarter Kelly, trading slower growth for a much wider margin against being wrong. The trap: sales material uses conviction language — “high-confidence opportunity” — to justify concentration that Kelly itself would reject once honest uncertainty about the edge is admitted. A pitch that names an opportunity but no sizing discipline has answered the easy question and skipped the one that decides survival.
Market regime #
The prevailing market weather — the combination of growth, inflation, and money conditions that decides how every asset behaves.
A market regime is the prevailing combination of growth, inflation, and money conditions under which markets operate — the financial weather. The same asset behaves differently in different regimes: a government bond cushions a growth scare but deepens losses in an inflation shock, as 2022 demonstrated when equities and bonds fell together. Regime-switching models (Hamilton, 1989) formalise the idea: markets occupy hidden states that persist for months or years, then change abruptly rather than drifting. The trap: products are marketed on track records earned in one regime — usually a long calm one — and presented as if the results were a permanent property of the strategy. A pitch that never states which regimes its record spans is describing summer clothing without mentioning the existence of winter.
Net liquidity #
A public estimate of how much central-bank money is actually free to chase assets, after subtracting the cash parked where markets cannot use it.
Net liquidity is the money actually available to financial markets after subtracting cash the system has parked out of reach — in the US case, roughly the central bank’s balance sheet minus the government’s cash account minus overnight reverse-repo balances. All three inputs are public; no terminal is required. The measure matters because liquidity changes tend to lead risk-asset prices by around a year (Howell’s global-liquidity framework): the tide moves before the boats do, through the balance-sheet capacity of the dealers who intermediate every market. Singapore adds a local wrinkle — its central bank steers through the exchange rate rather than a policy rate, so a strengthening Singapore dollar is itself a tightening signal. The trap: “liquidity” is invoked vaguely in marketing to justify any bullish story. The honest version is a number with a formula, a direction, and a lag — not a vibe.
Position sizing #
The decision of how much to hold, kept separate from what to hold — the discipline that determines whether one wrong idea can end the portfolio.
Position sizing is the decision of how much capital a single idea receives — deliberately separated from the decision of what to hold or which direction to lean. The separation matters because ruin is a sizing phenomenon: a correct thesis held at reckless size can still destroy a portfolio through one drawdown it cannot survive, while a wrong thesis at modest size is recoverable tuition. Practitioner discipline converges on the same rules: cap any single position, start small and add only as independent pieces of evidence agree, and hold a reserve that is never deployed at any level of conviction. The trap: sales material sells direction and conviction exclusively, because stories are about what to own, never about how much. A pitch that supplies certainty but no sizing rule has answered the entertaining question and omitted the one that decides survival.
Reaction function #
The rule connecting what a central bank sees to what it does — the mapping markets actually trade, since prices move on the expected path of policy, not today's setting.
A reaction function is the implicit rule connecting the data a central bank observes to the action it takes — how much inflation triggers how much tightening, how much stress triggers rescue. Markets trade the expected path of that rule, not the current policy setting, which is why a sentence can move trillions: in 2013, testimony merely mentioning a slower pace of bond-buying repriced ten-year yields by roughly a full percentage point within two months, though no policy had changed. Reading a 2026 rate decision is therefore two readings — what was done, and what the statement implies about the rule itself. The trap: commentary sells every communication wobble as a regime change. The useful distinction is between a communication shock (expectations repricing, usually violent and short-lived) and a policy shock (actual money conditions changing). Only the second changes the weather.
Real yield #
The return left after inflation — the only version of yield that measures purchasing power rather than a number on a statement.
Real yield is the return on an asset after subtracting inflation — the growth of purchasing power, not of the number on the statement. A deposit paying three percent during four-percent inflation has a real yield of roughly minus one percent: the balance rises while its buying power falls. This is why cash is not zero-risk; it is a small, certain, invisible loss in most years, and the reason “safe” is properly defined as matched to when the money is needed, not left in currency indefinitely. Real yields also steer asset prices broadly: when they rise, long-duration assets — bonds, growth stocks, property — reprice downward together. The trap: savings and income products are marketed on nominal rates in exactly the environments where inflation makes the real rate negative. Any yield quoted without the inflation rate beside it is only half a number.
Rebalancing #
Periodically trading a portfolio back to its target weights — trimming what has grown and restoring what has shrunk, by rule rather than by mood.
Rebalancing is the periodic resetting of a portfolio back to its target weights: positions that have grown beyond plan are trimmed, positions that have shrunk are restored. Its primary product is risk control. Without it, a portfolio drifts toward whatever has recently performed best, so the holder becomes most concentrated in the most expensive assets precisely at the top — an untouched 60/40 portfolio entering 2022 had drifted heavily toward equities after years of gains, maximum exposure at the worst moment. The mechanical effect of trimming winners and restoring laggards adds a modest contrarian return in range-bound markets and costs a little in long trends; the two roughly wash. The trap: rebalancing is sometimes marketed as a return-boosting bonus or a proprietary edge. It is neither. It is a discipline that keeps risk at the level chosen in calm, executed by rule when mood argues otherwise.
Risk-adjusted return #
Return measured against the risk endured to earn it — because return alone is only half a number.
Risk-adjusted return measures what an investment earned relative to the risk endured to earn it — through the Sharpe ratio, drawdown-based measures, or simply pairing every return with its worst loss. The premise: return alone is half a number. Twelve percent earned with mild swings and shallow losses is a different product from twelve percent earned through a sixty-percent drawdown, even though the raw figures match — the second demands a holder who never needs the money mid-journey and never loses nerve at the bottom. The trap: performance marketing compares raw returns against a benchmark while staying silent on the risk that produced them. “Beat the index” with triple the volatility and quadruple the drawdown is a loss dressed as a win. The completing questions are always the same pair: what was the worst fall, and how long did the recovery take?
Survivorship bias #
The distortion that comes from measuring only the funds and strategies that survived — the failures quietly vanish from the average.
Survivorship bias is the distortion created when performance statistics include only the funds, stocks, or strategies still alive at the end of the period — the failures were merged, closed, or delisted, and quietly left the sample. The effect is systematic flattery: an “average fund return” computed over survivors overstates what an investor choosing at the start would actually have earned, because some of the choices available then died along the way. The same bias animates single-trial success stories: for every publicised trader who compounded spectacularly, the unpublicised ones who ran the same approach and failed are invisible, so the method looks better than it is. The trap: fund families launch many products, close the laggards, and market the survivors’ records as the strategy’s expected outcome. The first question for any impressive average: how many started, and where are the ones that did not finish?
VaR (Value at Risk) #
The loss a portfolio should not exceed on 95 out of 100 normal days — a fence at the edge of normal that says nothing about what lies beyond.
Value at Risk (VaR) states the loss a portfolio should not exceed with a given confidence — “on 95 days out of 100, losses stay under this line.” It marks the fence at the edge of normal outcomes and says nothing about how far the ground drops beyond the fence. Two portfolios can share an identical VaR while one falls slightly past the line in a bad month and the other falls a hundred times further; VaR cannot tell them apart. In 2008, banks’ VaR models were roughly right about the 95th percentile and silent about the catastrophe past it. VaR also fails a basic coherence test (Artzner’s axioms): merging two portfolios can make measured VaR rise, so it lies when summed across a firm. The trap: pitches quote VaR as if it were the maximum possible loss. It is where the bad outcomes begin, not where they end.
Volatility #
How violently an investment's price swings around — usually quoted as a percentage per year.
Volatility measures how widely an investment’s returns swing around their average, usually quoted in percent per year. It is not just discomfort; it is a mathematical tax. What an investor actually compounds is approximately the average return minus half the volatility squared — so doubling the volatility quadruples the drag. An asset averaging twenty percent with forty-percent volatility compounds more slowly than one averaging fifteen percent with twenty-percent volatility; the flashier headline loses. The trap runs in both directions: sellers use scary volatility numbers to push investors into “smooth” products, and the smoothness is often fake — illiquid assets marked infrequently look calm because nobody has priced them lately, not because they are safe. A track record with no visible swings deserves more suspicion, not less.
Yield curve #
The line of interest rates a government pays across borrowing horizons, from months to decades — the market's posted price of time.
The yield curve is the line traced by a government’s borrowing rates across maturities, from short bills to long bonds — the market’s posted price of time. Its shape carries information: an upward slope is normal compensation for waiting; an inverted curve, with short rates above long, means markets expect policy to be forced lower, historically a recession-leaning signal. Its level moves everything else, because the government curve is the ruler every other asset is measured against — when it shifts, valuations reprice mechanically, hitting long-duration assets hardest. The trap: inversion gets sold as a countdown clock. The signal has preceded recessions by anywhere from months to years, and the lead time is unstable; marketing that converts a regime indicator into a dated prediction claims a precision the tool has never had. No one knows the lag in advance; the precision is claimed, never possessed.