Inside quant investing: alpha, beta and the strategies

Quant is short for quantitative. A quant investor decides what to buy and sell using numbers, written rules and statistical models, not a gut feeling or a visit to a company's headquarters. The rules are written down before any money is invested, they are tested on historical data, and then a computer applies them the same way every day.

Investing here means putting money into things like stocks, bonds and currencies in the hope they grow in value. A fund is a pool of money from many investors, run by professionals who make those decisions for everyone in the pool. A quant fund is a fund whose decisions come mainly from rules and models.

A famous chef cooks by taste and experience. She adjusts the salt by feel, notices when a tomato is a little sweet, and improvises when an ingredient is missing. Her results can be brilliant. But they depend on her, on her mood and on that day, and nobody else can copy them exactly.

Now imagine a food scientist who tests a recipe a million times. She changes the oven by one degree, the salt by one gram, and records how a thousand tasters rate each version. The final recipe is not exciting, but it works the same way for anyone who follows it, and she knows how often it fails.

Discretionary investing (where a person decides case by case) is the chef. Quant, also called systematic investing (where rules decide), is the tested recipe. Neither is automatically better. The chef can adapt to a kitchen fire. The recipe cannot, because it only knows what its tests contained. The recipe's strength is repeatability: it can run on thousands of ingredients at once, and it cannot get scared, greedy or tired.

Real funds sit along a line. At one end, a person makes every call. At the other, a computer does, and people only build and monitor the system. Many funds sit in between.

Here is a toy rule, using made-up numbers. There are 10 stocks. The rule says: at the start of each month, buy the 3 stocks whose price rose the most over the past 12 months. This idea is called momentum (stocks that have been rising tend, on average, to keep rising for a while).

Suppose the 12-month gains are: A 40%, B 35%, C 30%, D 20%, E 15%, F 10%, G 5%, H 0%, I -10%, J -20%. The rule buys A, B and C, investing one third in each. Next month they return +3%, +1% and +2%, so the portfolio return is (3 + 1 + 2) / 3 = 2%. The other seven stocks average, say, +0.5%. The rule's edge that month is 2% - 0.5% = 1.5%.

One month proves nothing. A quant would run this rule over decades of past data and across thousands of stocks, and ask: how often does it work, by how much, and is that result bigger than luck could explain? The rule's value lies in answering those questions with numbers.

1. 1952Harry Markowitz publishes Portfolio Selection, the first rigorous maths of diversification: judge an investment by its risk and return together, and by how it moves with the others in the portfolio.
2. 1960sThe Capital Asset Pricing Model (CAPM) is developed by several economists, including William Sharpe. It links an asset's expected return to its sensitivity to the market. That sensitivity is later called beta.
3. 1962Mathematician Edward Thorp publishes Beat the Dealer, showing that counting cards in blackjack can give a player an edge. He later applies the same mindset to markets.
4. 1967Thorp and Sheen Kassouf publish Beat the Market, about spotting mispriced warrants (a warrant is a long-dated option issued by a company).
5. 1973Fischer Black and Myron Scholes publish a formula for pricing options, with Robert Merton's related work. Derivatives trading grows fast, and so does the use of computers in finance.
6. 1970s-80sThe first index funds appear, and early quantitative firms start building statistical stock-selection models. Renaissance Technologies is founded in 1982.
7. 1990sAcademics document 'factors' such as value and momentum (Fama and French, Jegadeesh and Titman). Quant equity funds multiply.
8. 2007In early August, many quant equity funds lose sharply within days, often called the 'quant quake'. The likely cause was crowding: many funds held similar positions, and forced selling by some pushed prices against all of them.
9. 2010s-todayCheaper computing, huge new datasets and machine learning (software that finds patterns in data without being given the rule) become standard tools, alongside the older statistical methods.

Notice the pattern: each step makes some piece of investing measurable. Risk became a number in 1952, market sensitivity in the 1960s, and option prices in 1973.

1. Find signals. A signal is a measurable clue that predicts future returns a little better than chance. Examples are momentum, cheapness relative to earnings, or the tone of company announcements. Most signals are weak, and the edge is usually small.
2. Build portfolios. Combine many signals into one view, then decide how much to hold of each asset while controlling risk, so that no single bet or theme can hurt too much.
3. Trade cheaply. Every trade has costs: commissions, the gap between buy and sell prices (the spread), and the price moving against you as you buy (market impact). A signal that earns 0.30% but costs 0.40% to capture is worth nothing.

Costs deserve one more example. Say a signal predicts a 0.30% gain on a trade. Buying and later selling costs 0.10% each way, so 0.20% in total. The net gain is 0.30% - 0.20% = 0.10%. Cut the costs in half by trading smarter and the net gain doubles to 0.20%. This is why quant firms invest so heavily in trading, not only in prediction.

• Not magic. A model cannot see the future. It estimates probabilities, and it is wrong often. A strategy that wins 52% of the time can still be a good business over thousands of bets.
• Not necessarily fast. High-frequency trading (holding positions for fractions of a second) is one corner. Many quant funds hold positions for weeks or months.
• Not necessarily artificial intelligence. Many successful models are plain statistics, such as regressions (fitting a line through data). Machine learning is one tool among several.
• Not human-free. People choose the data, design the tests, set the risk limits and decide when to switch a model off.

A return is the percentage change in the value of an investment over a period. If you put in $100 and it becomes $110 a year later, the return is (110 - 100) / 100 = 10%. Returns can be negative: $100 becoming $90 is a -10% return.

Risk is the chance that things go worse than hoped. The most common way to measure it is volatility, which is the typical size of the bumps in returns. If an investment returns 10% every single year, its volatility is zero. If it returns +30% one year and -10% the next, it is bumpy, and its volatility is high. As a rough guide, an annual volatility of 15% means that in about two years out of three, the return lands within 15 percentage points of the average.

A second risk measure is drawdown: the fall from a previous peak to a later low. If a fund rises to $100, drops to $60, then recovers, its maximum drawdown is (100 - 60) / 100 = 40%. Drawdown matters because it is what investors actually feel, and because climbing back is harder than falling: from $60 you need a gain of (100 - 60) / 60 = 66.7% to return to $100.

The market here means a broad basket of stocks, usually measured by an index such as the S&P 500 (an index tracking 500 large US companies). When people say 'the market is up 10%,' they mean that basket rose 10%.

Beta measures how strongly an investment moves when the market moves. A beta of 1.0 means it moves about one for one with the market. A beta of 1.2 means it tends to move 20% more. A beta of 0.5 means it moves about half as much. A beta of 0 means it does not systematically follow the market.

If you own a stock with a beta of 1, and the market rises, you profit without having shown any skill. That return came from being exposed to the market. Alpha is the return left over after you remove what the market exposure explains. In a simple form:

alpha = fund return - (beta x market return)

A more careful version also removes the risk-free rate (what you could earn with almost no risk, such as short-term government bills). Say it is 2%. The fund's expected return from the market is then 2% + 0.8 x (10% - 2%) = 2% + 6.4% = 8.4%, and alpha = 14% - 8.4% = 5.6%. The idea is the same: alpha is what is left once the market's contribution is taken out. Try your own numbers below.

Beta is available to anyone for almost nothing. You can buy a fund that holds the whole market for a fee often well under 0.1% a year. Because it is so cheap and so easy, there is no reason to pay a high fee for it.

Alpha is rare because markets are competitive. Thousands of smart people look for mispriced assets. When someone finds an edge, others copy it, and the edge shrinks. Also, alpha is zero-sum before costs in a specific sense: if one investor beats the market, others holding the rest must on average trail it by the same amount. After fees, the average active investor falls behind. This is why paying high fees for beta that is disguised as alpha is the most common mistake.

The Sharpe ratio divides the return above the risk-free rate (the excess return) by volatility. It answers: how much extra return did I get for each unit of risk?

A rough guide: a Sharpe of 0.3 to 0.5 is typical for a broad stock market over long periods, around 1 is very good for a real fund, and anything much above 2 deserves suspicion unless the strategy has very short holding periods. Past Sharpe ratios also say little about the future.

The information ratio is like Sharpe, but it compares a fund against a benchmark (a yardstick such as an index) rather than against cash. It equals the active return (the fund's return minus the benchmark's) divided by tracking error (the volatility of that difference). If a fund beats its benchmark by 3% a year, and the gap swings with a volatility of 4%, then the information ratio is 3 / 4 = 0.75.

Correlation measures how two things move together, on a scale from -1 to +1. At +1 they move in lockstep. At 0 they are unrelated. At -1 they move in opposite directions. A fund with a correlation near zero to the market is a diversifier: it adds return that arrives at different times than your stock holdings. Quant funds often aim for low correlation to the market, which is why correlation is in this chapter.

A factor model explains an investment's return as the sum of exposures to a few common drivers. Four are famous:

• Market: the overall rise and fall of stocks (beta).
• Size: smaller companies have historically tended to earn somewhat different returns than large ones.
• Value: stocks that are cheap relative to measures such as earnings or book value have historically tended to beat expensive ones over long periods, though with long bad spells.
• Momentum: recent winners tend, on average, to keep outperforming for a while.

Smart beta (also called factor investing) packages these exposures into cheap, rule-based funds. Smart beta is closer to beta than to alpha: the factor returns are widely known and sold at low cost. True alpha, in the strict sense, is what remains after all the known factors are subtracted. Be careful with the labels: marketing sometimes calls a factor exposure alpha. A fair test is to ask whether the return can be reproduced by a cheap, simple factor fund. If yes, it is not alpha.

People use the words 'fund' loosely, so start with plain definitions.

• An index fund holds all the stocks (or bonds) in an index, in the same proportions, and aims to match that index. It makes money by earning the market's return, minus a tiny fee. Nobody picks stocks. An ETF (exchange-traded fund) is a fund whose shares trade on a stock exchange all day. Many ETFs are index funds.
• A mutual fund pools many investors' money into a portfolio. Investors buy and sell with the fund at a price set once per day, called net asset value (NAV, the value of the fund's holdings per share). Many mutual funds are active, meaning a manager tries to beat an index by picking investments. They typically hold mostly long positions (they own the assets), with limited or no borrowing.
• A hedge fund is a privately offered pool of money, open to limited kinds of investors, that is allowed to use tools ordinary mutual funds mostly avoid: shorting, borrowing and derivatives. It aims to make money in various market conditions, and charges higher fees.
• A quant fund is any of these, run mostly by rules and models. Most well-known ones are hedge funds, but quant mutual funds and quant ETFs (such as rule-based factor funds) exist too.

To hedge is to reduce a risk by taking an offsetting position. A farmer who locks in a price for next year's wheat is hedging. In investing, a hedge fund historically meant one that hedged out the market's ups and downs by holding both long and short positions.

A long position is owning something, which profits when the price rises. A short position is borrowing an asset, selling it now, and later buying it back to return it. A short position profits when the price falls. Example: you borrow a share and sell it for $50. The price drops to $40, you buy it back for $40 and return it, and you keep $10. If instead it rises to $60, you lose $10. Unlike buying, where the most you can lose is what you paid, a short's loss has no ceiling because a price can keep rising.

A fund that keeps its long and short exposure about equal is market-neutral: its beta is near zero. That is the cleanest way to try for pure alpha. Many quant hedge funds run this way, holding hundreds or thousands of small longs and shorts. Leverage (using borrowed money so that positions are bigger than the fund's own capital) is common here: because the net bet is small, the fund borrows to scale up the small expected gains.

These are typical patterns, not rules. Individual funds differ widely, so read the actual documents of any fund before drawing conclusions.

A classic hedge fund fee is called '2 and 20.' The 2 is a management fee: 2% of the money invested, charged every year whether the fund makes money or not. The 20 is a performance fee: 20% of the profits.

Many funds add a hurdle (no performance fee until returns pass a set level) and a high-water mark (no performance fee on regaining past losses). Real terms vary, and some of the most sought-after funds have charged more, and some newer funds charge less. The point is that fees take a large share of the profit, so a hedge fund must deliver considerable gross alpha just for you to come out ahead of a cheap index fund.

A multi-manager fund, sometimes called a pod shop, is one large firm that hands money to many small independent teams, called pods. Each pod runs its own strategy, often market-neutral, under strict risk limits. If a pod loses a set amount, it is typically cut or closed. The firm combines the pods so that the whole fund has steady returns and low correlation to the market. Some pods are quant and some are discretionary.

A pass-through fee means the fund charges investors for its operating costs directly, on top of or instead of a traditional management fee. These costs can include pay for the pod managers and analysts, data, technology and trading. Because they are passed through, a fund's total cost to investors can be well above the headline management fee, so investors should ask for the all-in cost. As an invented example, if a fund charges a 0% management fee but passes through costs of 6% of assets, you pay 6% before any performance fee.

Liquidity means how quickly you can turn an investment into cash at a fair price. A mutual fund or ETF is liquid: you can sell on any trading day. Hedge funds are usually not. Typical terms include a lockup (a period, often a year, when you cannot withdraw), redemption windows (you can withdraw only monthly, quarterly or annually) and notice periods (you must give, say, 30 to 90 days' warning). Some funds can also impose a gate, limiting how much can leave at once, in a crisis.

Many hedge funds are open only to accredited investors. In the United States, one way to qualify is to have a net worth over $1 million excluding your home, or income above $200,000 a year ($300,000 with a spouse). Others require even more, such as qualified purchaser status (generally $5 million in investments). The rules differ by country. On top of that, the very best quant funds have limited capacity, so some are closed to new money or to outsiders altogether. Capacity matters because each trade moves prices a little, and more assets mean more price impact.

Some firms offer quant ideas in mutual funds and ETFs, such as factor funds that tilt toward value, momentum or quality, and 'alternative' funds that run long/short strategies. They are cheaper and more liquid than hedge funds, but they have more limits on leverage and shorting, so their results can differ from those of private quant funds. Fees and past results vary widely, so each fund must be judged on its own documents.

Quant equity strategies start with a simple move: take thousands of stocks, measure the same trait for each one, and rank them. A trait that has been linked, over long periods, to higher or lower future returns is called a factor. Almost everything in this chapter is a variation on four steps: rank, bet, rebalance, repeat.

Every strategy also needs a reason it might earn money. There are four honest reasons. A risk premium is pay for carrying a risk that other people avoid. A behavioural error is a mistake people make again and again, such as reacting too slowly to news. Structural or forced flows are trades someone must make for rule-based reasons (an index fund must buy what enters its index, whatever the price). Information speed means knowing something, or acting on it, a little before the crowd. Each strategy below is labelled with its likely source. Nobody knows for sure which source is the real one, and that uncertainty is itself a risk.

• Value: favour stocks that are cheap compared with what the company earns or owns, for example a low price relative to yearly profit, or relative to book value (assets minus debts on the accounts). Likely source: part risk premium (cheap firms are often troubled) and part behavioural error (people extrapolate bad news too far). Held for months to years. It loses badly for long stretches when expensive growth stocks keep winning.
• Momentum: favour stocks that rose over roughly the past 12 months (usually skipping the latest month) and short those that fell. Likely source: behavioural, since investors under-react to news and then herd. Held for a few months. It can work for years, then crash sharply when beaten-down stocks suddenly rebound.
• Quality: favour firms with steady profits, low debt and stable earnings. Likely source: behavioural (investors overpay for exciting, low-quality stories) with some risk-premium flavour. Held for months to years. It can lag when low-quality stocks rally.
• Low volatility: favour stocks whose prices swing less. Simple theory says riskier stocks should earn more, yet calm stocks have often matched them. Likely source: behavioural (people like lottery-style stocks) plus structural (managers who cannot use leverage reach for riskier stocks to beat an index). Held for months. It lags in strong bull markets and can become expensive and crowded.
• Size: small companies have, on average, earned more than large ones. Likely source: risk premium (small stocks are harder to trade and fail more often). The premium has often been weak and unreliable, and small stocks are costly to trade.
• Profitability: favour firms that earn a lot of gross profit (sales minus the direct cost of making the product) relative to their assets. A close cousin of quality. Likely source: behavioural, since the market may undervalue durable profits. It can fade once profitable firms are fully priced.

1. Pick the universe: for example the 1,000 largest listed stocks, so trading stays cheap.
2. Compute the signal for each stock, say profit divided by price (called the earnings yield).
3. Clean it: cap extreme values and compare firms within their own industry, so the portfolio is not secretly a bet on one sector.
4. Rank all stocks from best to worst and cut the list into ten equal groups, called deciles.
5. Buy the top decile (the long side) and short the bottom decile (the short side). Shorting means borrowing shares and selling them, hoping to buy them back cheaper.
6. Size the two sides to equal dollar amounts and check that overall market exposure is near zero. This is market neutral: the portfolio should not care whether the market rises or falls.
7. Rebalance on a schedule (monthly is common), subtract trading costs, and repeat.

The long side: A rose 3%, so 3% of $100 is +$3. The short side: J fell 2%, so the short gains 2% of $100, which is +$2. Total: +$5 on $100 long and $100 short. The average stock rose only 0.7%, and the strategy did not need it. Now imagine a bad month: everything falls, A drops 8% and J drops 12%. The long loses $8, the short gains $12, and the net is +$4 even though the market fell. The reverse also holds: if A falls more than J, you lose even in a rising market. Real portfolios hold hundreds of names per side, so one wrong stock matters little.

A long-only fund cannot act on its worst-ranked stocks except by not owning them, so part of the signal is wasted. A market-neutral fund uses both sides, but its raw profit is small (the gap between winners and losers), so it often uses leverage, meaning borrowed money, to make the return worth having. That magnifies losses too. A short squeeze is the shorts' nightmare: a heavily shorted stock rises, short sellers are forced to buy shares back, and their buying pushes the price up further. A long loses at most 100% of what was invested, but a short can lose more than 100%, because a price has no ceiling.

Statistical arbitrage, often shortened to stat arb, is a large basket of small bets that relative prices will return to normal. The simplest version is a pair: two similar stocks, in the same industry with similar businesses, that usually move together. When one runs ahead of the other, you short the one that ran and buy the other, expecting the gap to close. It is called arbitrage, but it is not risk-free. It wins on average and can lose.

Real stat arb runs hundreds or thousands of such bets at once. Models first strip out the part of each stock's move explained by the market, its sector and common factors, leaving a residual (the stock's own move), and then bet that residuals revert. Holding periods run from days to a few weeks. Likely source: mostly structural (the fund is paid for providing liquidity to impatient sellers who push prices briefly away from fair value) plus behavioural overreaction. Capacity is moderate. It is very crowded, because many funds run similar models. In August 2007, many stat arb funds were forced to cut positions at the same time and lost heavily over a few days, with no news about the underlying stocks.

• Days to one month: recent losers tend to bounce and recent winners give a little back (short-term reversal). Likely cause: impatient or forced sellers move the price, then liquidity providers earn money as it returns.
• About 3 to 12 months: winners tend to keep winning (momentum). Likely cause: slow reaction to news and herding.
• Three to five years: the biggest long-run winners and losers tend to flip back (long-term reversal), a close cousin of value.

These do not contradict each other: different forces dominate on different time scales. This is why the momentum signal skips the latest month, so it is not contaminated by short-term reversal. Quant funds usually blend several signals, but the signals can pull against one another.

When a company reports profit, the surprise is the gap between the reported figure and what analysts expected. Prices often do not adjust in one jump. After a good surprise, stocks have tended to keep drifting up for weeks or months, and after a bad one, down. This is called post-earnings-announcement drift. Likely source: behavioural (under-reaction) and information speed (news spreads slowly through the market). Illustration: analysts expect 1.00 per share, the company reports 1.20, a +20% surprise. The stock jumps 3% that day, then drifts a further 2% over the next 60 trading days. A strategy buys the biggest positive surprises and shorts the biggest negative ones after the report, holding for weeks to a few months. Risks: much of the move now happens in the first minutes, where fast traders take it, and drift is usually smaller in large, well-covered stocks.

Factor timing means trying to hold a factor only when it is about to do well. Factors do suffer long droughts, but predicting them is hard and the evidence is mixed, so most funds hold several factors for diversification instead. Factor crowding means so many funds hold the same stocks that the trade becomes expensive and fragile. Research finds that published factor premiums tend to shrink afterwards. When one crowded fund must sell (losses, investor withdrawals), prices fall and hurt every other holder, who then sell too. This is a crowded unwind. The sharp momentum crash of 2009, when the stocks that had fallen most in the financial crisis rebounded fastest, is a textbook case.

The strategies in this chapter mostly trade futures: contracts to buy or sell something at a set price on a future date, covering stock indexes, government bonds, currencies and commodities. Futures need only a small deposit called margin, so a fund can hold large positions, and it can sell short as easily as it can buy.

Time-series momentum looks at each market's own past. If a market has been rising, buy; if it has been falling, sell short. It differs from the stock momentum of chapter 04, which ranks stocks against each other. Funds that do this are called CTAs (commodity trading advisors) or managed futures funds. Likely source: behavioural (investors under-react to news and herd) plus structural (hedgers such as farmers and airlines trade for protection and may pay trend followers to take the other side). Positions are held from days to several months.

The long trade: bought at 104 on day 5, exited at 108 on day 9, a gain of 4 per unit. The price peaked at 114, so the rule gave back 6 from the top. The short trade opened at 108 and is up 3 per unit at 105. Now picture a market with no trend: price wobbles 104, 103, 105, 102. The rule buys at 104 and sells at 103 (-1), then shorts at 103 and buys back at 105 (-2). That is -3 with no trend at all. This is called whipsaw. A trend follower expects many small losses and a few large wins.

Positioning by volatility. Volatility is how much a price typically swings. A fund does not put the same dollars in every market, because a calm market and a wild one would then carry very different risk. Instead it sizes each position so it contributes similar risk. If market A swings about 10% a year and market B about 20%, a fund targeting 10% for each holds a position in A equal to 10 / 10 = 100% of capital and in B equal to 10 / 20 = 50% of capital. Both then add about the same risk. When a market becomes more volatile, the position is cut automatically.

Why it tends to do well in crises and badly in choppy markets. In a long, steady slide (for example a prolonged fall in stocks, or a year when bond prices fall as interest rates rise), the rule flips to short and profits, and it has no built-in bias toward owning stocks. In a sideways market or a sharp reversal after a long trend, signals flip too late and the fund pays whipsaw losses. Trend followers generally did well in 2022, a year of sustained trends in rates and currencies, but a fast V-shaped rebound can hurt. Capacity is large in liquid futures, though very large funds move prices when they trade. Crowding is moderate to high: many funds use similar look-back windows (one to twelve months), so many hold the same direction at the same time.

Global macro means betting on country-level themes such as interest rates, inflation, growth and currencies. Discretionary macro relies on a manager's judgement. Systematic macro turns the same ideas into rules that read data (inflation, growth, central bank policy, how crowded other investors' positions are) and set positions across currencies, bonds, stock indexes and commodities. Likely source: a blend of risk premium (being paid to hold unloved assets), behavioural error (markets are slow to price in a changing economy) and information speed (processing data quickly). Holding period is weeks to months. Capacity is large. The main risk is a regime change, a shift in how the economy behaves, with few clean examples (recessions are rare) for a model to learn from.

Carry is the return you earn if prices stay exactly where they are. Currency carry means borrowing in a currency with low interest rates and lending in one with high rates. Bond carry earns the extra yield of longer bonds. Commodity carry comes from roll yield, which is the gain or loss when a futures contract is rolled forward. The likely source is a risk premium (the high-rate currency is often one that can crash suddenly) mixed with structural flows (some big players hedge and do not care about price). Holding period is weeks to months.

Carry does well in calm markets and loses in crises, when investors suddenly flee high-yielding assets. Capacity is large. Crowding is high, since carry is one of the best-known ideas in finance and shared by many funds.

A classic 60/40 portfolio puts 60% in stocks and 40% in bonds. The dollars are balanced, the risk is not: stocks swing far more than bonds, so stocks drive almost all the ups and downs. Risk parity instead sizes each asset so it contributes roughly equal risk. Because bonds are calmer, they get a bigger weight, and the whole portfolio is then leveraged (scaled up with borrowed money) so it still aims for a useful return. Likely source: risk premium (diversifying across assets that each earn a premium) plus a structural reason (many investors cannot borrow, so they overpay for risky assets and leave safer assets cheaper).

The 2022 lesson, in words. The plan relies on bonds acting as a cushion when stocks fall. In 2022, with inflation high and interest rates rising fast, stocks and bonds fell together. The big, leveraged bond position, which was supposed to be the safe leg, lost money at the same time as stocks. Diversification that rests on a relationship between assets works only while the relationship holds. Risk parity capacity is very large, and the idea is shared by many large funds, so crowding is moderate to high.

Insurers earn a steady premium and pay out rarely and heavily. Selling volatility works the same way: the fund sells options or volatility futures, collects premium, and keeps it if markets stay calm. Likely source: risk premium, because investors have tended to pay more for protection than the protection later cost. It looks like a steady, high-return strategy, until it is not.

Merger arbitrage: when one company agrees to buy another, the target's shares usually trade slightly below the offer, because the deal might fail. The fund buys the target and collects the gap if the deal closes. Likely source: risk premium for bearing deal-break risk and structural flows (holders who want certainty sell). Holding period is weeks to months. Convertible arbitrage: a convertible bond is a bond that can be swapped into shares, so it is a bond plus an option. The fund buys the convertible, shorts some of the underlying shares to cancel the stock-price risk, and earns the bond's interest plus gains from trading the option. Likely source: structural (issuers and sellers who want cash, so the option can be cheap). Its main risks are widening credit spreads, lenders cutting leverage, and crowding.

The earlier chapters were about what to hold. This chapter is about how fast you act, what it costs to act, and what tools find signals. It covers market making, high-frequency trading, trading costs, alternative data and machine learning. A recurring lesson: a good idea on paper can be ruined by costs, and a strategy that depends on speed has a small, expensive, crowded playing field.

A market maker posts two prices at once: a bid (the price at which it will buy) and an ask (the price at which it will sell). The gap between them is the bid-ask spread. Buyers and sellers who want to trade right now pay that gap for convenience. A market maker stands in the middle, buying at the bid and selling at the ask, and keeps the difference. Likely source: structural (it is paid for providing immediacy), with information speed as a tool. It holds positions for seconds to minutes, and ends the day with little or nothing.

Market makers manage inventory by skewing quotes. If they have bought too much, they lower both prices a little to attract sellers less and buyers more. They also fear adverse selection, which means trading against someone who knows more. If a well-informed seller hits your bid just before bad news, you hold shares that are about to drop. The spread is partly payment for that risk. Margins are thin, competition is intense, and capacity per stock is small, but a firm can run it across thousands of instruments. The idea is shared by many firms, and the main risk is a sudden jump in prices that leaves inventory underwater.

High-frequency trading, or HFT, is not one strategy. It is a style of automated trading that sends very many orders, holds positions for fractions of a second to minutes, and usually ends the day flat. Inside it you find market making, small price differences between venues or between related instruments, and very short-term prediction from the order flow. Latency means delay: the time between an event and your reaction, measured in microseconds (millionths of a second).

• What it is: speed as a competitive edge. A fast firm sees a price change on one venue and adjusts its quotes on another before slower firms do.
• What it is not: it does not see your order before you send it. Using advance knowledge of a client's order to trade ahead of it is illegal front-running. Fast firms react to prices that are already public, only sooner than you can.
• Co-location means placing a firm's servers in the same data centre as the exchange's, so signals travel the shortest distance.
• Microwave links: light in glass fibre travels at roughly two thirds of its speed in air, and fibre routes are not straight. Towers sending microwave signals through the air along a near-straight line between distant trading centres can shave off milliseconds. Firms pay heavily for that edge.

The effect on the public is debated. Quoted spreads for everyday investors are far narrower than in the pre-electronic era, but critics point to episodes of extreme moves, such as the sudden market drop on 6 May 2010 (the flash crash). The business is winner-takes-most: the fastest firm wins the trade, the rest get nothing. Capacity is small because there is only so much spread to harvest, technology costs are high, and the edge is crowded and decays whenever a rival becomes slightly faster.

Execution is how a trade is actually done. Slippage is the difference between the price you expected and the price you got. Market impact is the part of slippage caused by your own order: if you buy a lot, you push the price up against yourself. Impact generally grows with order size, but more slowly than one for one; a common rule of thumb is roughly the square root of size. Transaction costs add up four things: commissions, the spread, impact, and for shorts the fee for borrowing shares.

Execution algorithms cut those costs by slicing a big order into small pieces. TWAP (time-weighted average price) spreads the order evenly over a time window: to buy 100,000 shares over five hours, it buys 20,000 each hour. VWAP (volume-weighted average price) follows the market's usual volume pattern: if 10% of the day's typical volume trades in a given hour, it buys 10% of the order, or 10,000 shares, in that hour. The aim is to trade like the crowd and avoid moving the price. Both are blunt tools: they try to be unnoticed, not to be clever, and a patient order risks the price moving away while waiting.

Alternative data means information from outside the usual financial statements and prices. The idea is to learn something about a company or the economy before the official numbers arrive. Examples: satellite images that count cars in shopping-centre car parks or measure the fill level of oil tanks; anonymised card spending that shows a retailer's sales before it reports; web traffic and app downloads; job postings that show a firm's hiring; and text from news, company filings and earnings-call transcripts. Natural-language models read that text and score it, for example as positive or negative in tone, from simple word lists to modern large language models.

Edge source: mostly information speed, and it is perishable. Once many funds buy the same dataset, the signal gets priced in, and sellers of data often sell to many clients. Risks: data must be legally and ethically sourced (trading on illegally obtained inside information is a crime), samples can be biased, and the data can be costly. Capacity is moderate.

Machine learning means algorithms that find patterns in data on their own. They have transformed image and language tasks. Finance is a harder setting for three reasons.

• Low signal-to-noise: price changes are mostly random noise, and the predictable part is tiny. A model that explains even a small fraction of the next month's return variation can be valuable, and a model that seems to explain more is probably fooling itself.
• Little data: 20 years of monthly returns is just 240 observations per stock, far fewer than the millions of images that vision models learn from. Rare events like crises appear a handful of times.
• Regime change: relationships shift as the economy, rules and other traders change, so a pattern learned from the past may vanish.

Overfitting is the central danger: a model that memorises noise looks excellent in testing and fails live. A simple illustration: test 100 random signals, each with a 5% chance of looking significant by luck, and you expect 100 x 0.05 = 5 false winners. Researchers who try thousands of models and report the best one are almost sure to find luck. The defence is strict out-of-sample testing (testing on data the model never saw), simple models, and scepticism.

Where it helps: combining many weak signals, including non-linear interactions; reading text; predicting trading costs and how much a market will move for a given order; cleaning data; and risk modelling. The difference between prediction and trading is large. A forecast is only a number. A trade also needs size, costs, risk limits and capacity. Suppose a model predicts a stock will rise 0.10% tomorrow and is right 52% of the time, but the round trip costs 0.15%. Expected result per trade: 0.10% - 0.15% = -0.05%. A correct prediction still loses money after costs. Edge source: mostly information, through better processing, not better knowledge. Crowding is rising, since many firms hire similar talent and use similar tools.

A quant fund is an investment firm that decides what to buy and sell using rules written as computer code, tested on historical data, instead of a manager's gut feeling. Every serious quant firm runs some version of the same assembly line, and each stage can fail in its own way.

1. Data: collect prices, company accounts and other information.
2. Cleaning: fix errors and line the data up in time.
3. Signal research: find a pattern that predicts returns.
4. Backtest: replay the idea on the past.
5. Portfolio construction: turn predictions into actual position sizes.
6. Risk management: keep any one mistake survivable.
7. Execution: place the trades cheaply.
8. Monitoring: watch live results and retire ideas that stop working.

There are three broad kinds of data. Price data is what traded and when: open, high, low, close and volume. Fundamental data comes from company accounts: sales, profit, debt. Alternative data is anything else that might hint at the future, such as shipping records, web traffic, satellite images of car parks or the text of earnings calls.

Raw data is messy. A stock split makes a price look as if it halved overnight when nothing happened. A company changes its name or ticker. A vendor records a price of 0 for a day with no trade. Accounts are restated months later. Cleaning means fixing all of this, and above all storing data point-in-time: the version of the number that was actually known on that date, not the corrected version published later.

A signal is a number computed for each asset each day that is supposed to predict its future return. Example: the return over the past 12 months, skipping the latest month, is a classic momentum signal. Another is the ratio of a company's profit to its share price, a classic value signal. Researchers start from an idea, ask why it should work (a behavioural bias, a risk investors demand pay for, a constraint that forces others to trade badly) and then test it.

A signal with a reason is far more trustworthy than a bare pattern. 'Stocks whose tickers start with a vowel did well' can be found in data, but it has no reason, so it will not last.

A signal says which assets look good. A portfolio says how much money goes into each. This step has four parts. The expected return is the forecast for each asset, built by blending the signals. The risk model estimates how volatile each asset is and how assets move together. The optimiser is a program that searches for the position sizes giving the best forecast return for a given level of risk. The constraints are the rules it must respect: a cap on the size of any position, limits on exposure to one industry or country, a ceiling on leverage (borrowed money) and a limit on how much it may trade.

Risk management wraps around all of this: limits on losses, market exposure, leverage and crowded positions, often with the power to cut positions whatever the signals say. Execution then carries out the trades. Trading is not free: you pay the gap between the buying and selling price (the spread) and you move the price against yourself when you trade in size (market impact).

Research is a loop: idea, test, review, small live trial, scale up, monitor, repeat. Most ideas fail early, and good firms keep a record of every test they ran. The team usually combines researchers (often with PhDs in maths, physics, statistics or computer science) who find signals, engineers who build data systems and trading software, traders who manage execution and market structure, and a risk group that can say no. Renaissance had roughly 310 employees in 2021, about 150 of them researchers and programmers, while D. E. Shaw has around 2,500 employees.

The costs are mostly fixed and large: salaries, computing power, fast connections to exchanges and data subscriptions. This favours big firms.

• Overfitting: tuning a model until it fits the noise of the past instead of a real pattern.
• Lookahead bias: using information in a test that would not have been available at the moment of the decision.
• Survivorship bias: testing only on companies that still exist today, ignoring those that failed or were dropped.
• Data snooping: running many tests on the same data and reporting only the winner.

The usual rule of thumb for 'significant' is a result so strong that it would occur by luck less than 5% of the time. That works for one test. If you run 200, you expect about 10 lucky winners. The fix is to raise the bar as the number of tests rises. A simple correction divides the 5% by the number of tests: 0.05 / 200 = 0.00025, which needs a t-statistic of about 3.7. Academics who study the huge number of published factors have argued, in broad terms, that the old threshold of 2 is too low and that something nearer 3 is more honest. Count every test you ran, including the failed ones.

Edges fade for two reasons. First, once a pattern is published or noticed, others trade it, and trading removes it. A 2016 Journal of Finance study by McLean and Pontiff examined 97 return predictors and found their returns were about 26% lower out of sample and about 58% lower after publication. Second, crowding: when many funds hold the same positions, selling by one pushes prices against all (Chapter 9).

Capacity is how much money a strategy can handle before its own trading eats the profit. Because market impact rises with size, returns per dollar fall as assets grow. This is why some of the most profitable funds stay small or turn money away (Chapter 8).

Firms that look similar from the outside earn their money in quite different ways, and that shapes who can invest and what you pay. Where sources disagree, ranges are given and flagged.

• Fees: the firm manages other people's money and charges a percentage. A hedge fund is a private investment pool for wealthy or institutional investors; the 'standard' fee was historically '2 and 20': 2% of assets each year plus 20% of profits, though the fee range is wide (roughly 1-4% and 10-50%) and rates have been falling since 2008.
• Trading its own capital: a proprietary (prop) trading firm has no outside clients and keeps all profits and losses.
• Market-making spread: a market maker quotes both a price to buy and a price to sell and earns the gap between them, thousands of times a day.

Renaissance was founded in 1978 as Monemetrics by Jim Simons and Howard Morgan and renamed in 1982. Its Medallion fund started in 1988. According to Gregory Zuckerman's 2019 book, Medallion returned about 66% a year before fees and about 39% after fees since 1988. Reported net returns include +55.9% in 1990, +98.2% in 2008 and +76% in 2020; even the worst year from 2001 to 2013 was +21% net.

Medallion has been closed to outsiders since 1993, open only to employees and their families, and the last outside investors were bought out in 2005. Renaissance also runs funds for outsiders, such as RIEF (created 2005) and RIDA. They have trailed Medallion by roughly 17-19 percentage points a year, per a 2020 report. RIEF reportedly fell 8.7% in August 2007, the month of the quant quake (Chapter 9). Firm-wide assets are quoted at roughly 75 billion dollars (a Bloomberg figure from November 2020) up to about 130 billion (April 2021), and the sources are inconsistent, so treat any single number loosely. Simons stepped down as chief executive in late 2009 and died in 2024.

Two Sigma was founded in 2001 by John Overdeck, David Siegel and Mark Pickard, with seed money from Paul Tudor Jones. Reported assets grew from about 8 billion dollars in November 2011 to over 50 billion in October 2017, about 58 billion in October 2020 and about 70 billion in 2025. Founder discord was disclosed in 2023, unauthorised model changes were reported in October 2023 (an SEC probe followed and clients were compensated), and in September 2025 US prosecutors indicted a former researcher, Jian Wu, over alleged client losses of about 165 million dollars. Carter Lyons and Scott Hoffman have been co-CEOs since 30 September 2024.

D. E. Shaw was founded in 1988 by David E. Shaw, a former Columbia computer science professor, and began investing in June 1989 with 28 million dollars. It reported 55 billion at 1 June 2021 (35 billion alternative, 20 billion long-only) and about 65 billion in 2025, with roughly 2,500 employees. An executive committee has run it since Shaw stepped back in 2002. It mixes systematic and discretionary strategies, so it is not purely quant.

AQR was founded in January 1998 by Cliff Asness, David Kabiller, John Liew and Robert Krail, all formerly of Goldman Sachs. It built its business on factor investing: systematic exposure to value, momentum and defensive stocks, carry, plus risk parity and managed futures. Assets are quoted from about 193 billion dollars in 2018 to 143 billion in 2020 and about 128 billion in 2025, and one source shows 99 billion for the end of 2023. These series disagree, so treat the numbers loosely. Profit reportedly fell 34% in 2018 and jobs were cut in January 2019.

A trend follower buys markets that have been rising and sells markets that have been falling, across futures in stocks, bonds, currencies and commodities, and cuts losses quickly. Man Group dates from 1783; AHL was founded by Cyril Adam, David Harding and Martin Leuck and acquired by Man between 1989 and 1994. Man Group reported about 227.6 billion dollars in assets for 2025 (whole group, not AHL alone). Winton was founded by David Harding after he left AHL and is also a trend follower.

Bridgewater was founded in 1975 by Ray Dalio. Its Pure Alpha fund trades global macro, and All Weather, launched in 1996, is a risk-parity portfolio that balances risk (not dollars) across stocks, bonds and other assets. Macro means trading economy-wide themes such as interest rates and currencies. Its assets are among the largest in the industry; a precise figure was not verified here.

Citadel was founded in 1990 by Ken Griffin, who owns about 85% of it, starting with 4.6 million dollars. It reported assets above 77 billion in January 2026, though another section of the same source gives 67 billion at the start of 2026, so roughly 67-77 billion. In 2008 its flagship funds fell 55% with leverage near 7 to 1 and a ten-month gate (a block on withdrawals), then gained 62% in 2009. In 2022 it reportedly earned about 16 billion for clients, a record in dollars. Citadel Securities is a legally separate market maker.

Millennium (founded 1989 by Izzy Englander) and Point72 (founded 2014 by Steve Cohen, from SAC) use the same idea: a platform hires many semi-autonomous portfolio manager teams (pods), gives each capital, enforces strict drawdown limits and cuts pods that lose too much. Most pods are discretionary, meaning humans choose trades, notably in equities, though each platform also has quantitative pods. Their assets are in the tens of billions of dollars. They are not 'quant funds' with one central model but risk-managed collections of many strategies, some of them quant.

The fee model is different too. In a pass-through model, investors pay the pods' costs (salaries, data, financing) on top of management and performance fees. Cambridge Associates reportedly put these pass-through charges at between 0.87% and 12.99%, with a mean of 5.93% and median of 4.53% (October 2025). Another source gives 3-10%. An advocacy-leaning estimate from March 2026 says investors keep about 41 cents of each gross dollar; treat that as a view.

A market maker stands ready to buy and sell, which gives other traders somewhere to trade at any time. It profits from the spread and from speed and scale, and tries to end each day holding little risk.

Jane Street was founded on 31 August 1999 by Tim Reynolds, Rob Granieri, Marc Gerstein and Michael Jenkins. It is a market maker and proprietary trader, is known for programming in OCaml, has no CEO, and has about 3,000 staff (2024). Its reported revenue was about 39.6 billion dollars for fiscal 2025, with net trading revenue of 10.1 billion and net profit of 6.9 billion in the second quarter of 2025. It reportedly accounted for around 41% of bond ETF volume in 2024. In July 2025 India's regulator SEBI alleged Bank Nifty manipulation. Jane Street escrowed 560 million dollars on 14 July, a ban was lifted on 21 July 2025, and the firm denies wrongdoing. Hudson River Trading, Virtu (publicly listed), Jump and Tower Research are similar kinds of firm; figures for them were not verified and are left out.

Quant strategies fail because the world changes, because too many people do the same thing, because leverage turns a small loss into a fatal one, or because the machinery breaks. Where sources disagree it is noted.

In early August 2007, many equity market-neutral quant funds lost money at once, in a market where stocks overall hardly moved. Market neutral means holding long and short positions meant to cancel out the market's direction. Researchers Khandani and Lo concluded the trouble started with the forced unwinding of one or more large quant portfolios, via a margin call (a lender demanding more collateral) or a decision to cut risk. Selling those positions pushed down the longs and pushed up the shorts, which hurt every other fund holding similar positions. Those funds then cut risk through stop-losses and deleveraging, and the loop fed itself. A follow-up study identifies unwinds on 1 August (10:45 to 11:30) and 6 August and a pullback by market makers from 8 August. Prices rebounded around 10 August. Renaissance's outside fund RIEF fell 8.7% in August 2007.

Long-Term Capital Management began trading on 24 February 1994 with about 1.01 billion dollars. By the start of 1998 it had about 4.7 billion of equity but had borrowed more than 124.5 billion, a leverage ratio above 25 to 1, with roughly 1.25 trillion dollars of notional positions. Notional is the face value of contracts, not the money at risk. It lost 4.6 billion dollars in under four months. Equity fell from 2.3 billion at the start of September to about 400 million by 25 September. On 23 September 1998 the New York Fed organised a recapitalisation of about 3.6 billion dollars from 14 firms, who took 90% of the fund.

On 6 May 2010, starting at 2:32 pm Eastern time, the Dow Jones index fell 998.5 points, about 9%, within minutes, and recovered most of it by around 3:07 pm. About 1 trillion dollars of market value was briefly erased. A joint SEC and CFTC report (30 September 2010) described a large sell program of 75,000 E-Mini futures contracts, worth about 4.1 billion dollars, by the fund manager Waddell & Reed as the trigger. Later, Sarao was arrested in April 2015 and charged with 22 counts including spoofing (placing orders you intend to cancel to mislead others); he was sentenced in January 2020 to one year of home confinement.

Knight Capital was a major US market maker. On 1 August 2012 (some accounts give 31 July, but it is widely cited as the morning of 1 August, Eastern time) a software release went wrong. New code was not deployed to one of eight servers, and a reused flag woke dormant, old test code known as Power Peg. For about 45 minutes the system sent unintended orders: about 4 million executions in 154 stocks, 397 million shares. The loss was about 440 million dollars before tax (some accounts say 460 million), more than the firm could absorb. Its stock fell 33% that day and about 75% the next. A rescue of about 400 million, led by Jefferies, came on 5 August, and Getco agreed to acquire the firm in December 2012.

Volatility is how much prices move. The VIX is an index of expected volatility in US stocks. In early February 2018 volatility jumped abruptly after a long calm. Products that profited when volatility stayed low, including an exchange-traded note that bet against the VIX, lost almost all of their value in a day or so and at least one was shut down by its issuer. Exact figures are not given here. The lesson is about shape of payoff: selling volatility earns small steady gains for a long time and then suffers a large loss all at once, like selling insurance without enough capital behind it.

Not all failures are sudden. Momentum, buying recent winners and selling recent losers, tends to earn a premium but can crash. Academic research documents severe momentum reversals in sharp market rebounds after crises, with 2009 the standard example. Factor investing also had a long drought from about 2018 into 2020, when value (buying cheap stocks) reportedly had its worst stretch in decades before a sharp recovery from November 2020. For AQR, profit reportedly fell 34% in 2018 and jobs were cut in January 2019, and its assets fell from about 193 billion dollars in 2018 to 143 billion in 2020 (the series is inconsistent across sources).

Trend following swings between feast and famine. The SG CTA Index reportedly gained 20.2% in 2022, a record, and the SG Trend Index about 27.3%. Then 2023 was reported as weak, 2024 as volatile and the first half of 2025 as weak. Funds that look alike also move together: the average pairwise correlation among SG Trend constituents was reportedly 0.78 in 2024, yet performance still ranged across about 15 points. Trend followers do well when markets move in sustained directions and badly when they whipsaw.

Alpha is the part of a return that is not explained by exposure to the market or known factors. Alpha decays because good ideas get copied. McLean and Pontiff (Journal of Finance, 2016) studied 97 predictors and found returns about 26% lower out of sample and about 58% lower after publication. Capacity limits make it worse: strategies that work with a small amount of money stop working when large. Model risk is the plain fact that models describe the past: they assume old relationships hold, and fail when the regime changes, as with LTCM in 1998.

Then comes what you can actually keep. A 2 and 20 fee is a large share of gross profit, and pass-through costs can add more. A study by Patton and Weller (2020) found that after costs, typical mutual funds earn low returns to value and none to momentum. Average hedge fund returns in 2025 were reportedly 11.8% (Goldman Sachs), while the broad US stock market returned somewhat more.

• Count your leverage. It decides whether a bad month is survivable.
• Ask who else holds the same trade, and how the exit looks.
• Operations are a risk: test code, add kill switches, monitor live.
• A model is a description of the past, not a law.
• Edges decay once found and shrink as assets grow.
• Judge a strategy on what you keep after all costs, not on gross returns.
• Be wary of payoffs that look smooth and calm. They may hide a rare large loss.

Quant investing means making investment decisions with rules and data instead of gut feeling. So can a person do it? Honestly: yes for the simple version, no for the version the famous funds run. The big firms have teams of PhDs, years of cleaned data, fast computers, cheap trading and thousands of positions. A person at home has none of that. The edge those firms find is small, and it survives only because they can trade it cheaply and at scale.

What a person can do is use the same ideas in a slower, simpler, cheaper form. The ideas that last longest (owning broad markets, diversifying, following a trend, rebalancing) are old, well studied and available at low cost. The rest of this chapter sorts what is realistic from what is marketing.

These are categories, not recommendations. Each has costs and ways to disappoint.

Three rules need no code and no special data. Each is a small version of what quant funds do.

Rebalancing means resetting your mix back to target after prices move. Example: you hold 10,000, with a target of 60% stocks and 40% bonds, so 6,000 and 4,000. Stocks rise 20% and bonds stay flat. Now stocks are 7,200 and bonds 4,000, total 11,200, and stocks are 7,200 / 11,200 = 64.3% of the pile. To reset, you want 60% x 11,200 = 6,720 in stocks, so you sell 480 of stocks and buy 480 of bonds. The rule forces you to sell some of what rose. Limit: in a long, strong run it holds you back, and it does nothing about a market that falls together.

Dollar-cost averaging means investing a fixed amount on a fixed schedule. Example: you put in 100 each month and the price per share is 10, then 8, then 5, then 10. You buy 10 + 12.5 + 20 + 10 = 52.5 shares for 400, so your average cost is 400 / 52.5 = about 7.62, below the simple average price of 8.25. Limit: it spreads out when you buy, but it does not raise expected returns. If the price just climbs, investing everything at the start would usually have done better. Its real value is that it keeps you investing by rule.

A moving-average trend filter means holding an asset only while its price is above its average of the last N days (often 200 days), and holding cash otherwise. Example: an index closes at 4,100 and its 200-day average is 4,000. The rule says hold. Next month it closes at 3,900 and the average is 3,980. The rule says move to cash. Limit: in choppy markets the price crosses the line again and again, and each switch can cost money and taxes (a whipsaw). The filter can lower deep losses in long bear markets, but it can also leave you in cash just before a rebound. There is no promise either way.

A fund pitch is a claim, and claims can be checked. Go through the list in order. Any 'no' or 'we can't share that' should slow you down.

1. Is the track record audited by a real, independent firm? A chart in a slide deck is not an audited record.
2. Is it out-of-sample? Results from after the strategy was built and run with real money matter more than results the builder fitted to old data.
3. Is it net of fees and costs? Ask for returns after all fees, trading costs and any performance fee.
4. How many strategies were tried before this one was picked? If 100 versions were tested and the best is shown, the result is partly luck.
5. What is the capacity? Many edges stop working when more money chases them. Ask how much the strategy can hold before returns fall.
6. What was the worst drawdown (the drop from a peak to the lowest point)? Could you live through that fall without selling?
7. What is the Sharpe ratio, and how does it compare with a plain index fund over the same years? A Sharpe ratio is return above cash divided by volatility. Beating a cheap index only on paper is not a win.
8. Is there key-person risk? If one founder leaves or dies, does the process continue?
9. What are the liquidity terms? Ask how often you can withdraw, how much notice you must give, and whether a 'gate' can block withdrawals.
10. What are all the fees? Management fee, performance fee, hurdle rate (a minimum return before the performance fee), and whether losses must be made back first (a high-water mark).
11. Who holds custody of the assets? An independent custodian and administrator, not the manager itself, should hold the money and calculate values.
12. Are there fraud red flags? Look for returns that are far too smooth, an edge nobody can explain or verify, secrecy used as a selling point, a tiny or unknown auditor, and pressure to decide fast.

• Guaranteed or 'risk-free' returns do not exist in markets. Walk away.
• Higher return always comes with higher risk somewhere. If the risk is not visible, find it before investing.
• If you cannot explain in two sentences where the profit comes from, do not put in money you cannot lose.
• Never be rushed. A real opportunity will still be there after you read the documents.
• Be wary of any pitch that leans on a famous name, a secret method, or 'AI' without showing evidence.
• Do not invest on a backtest alone, however pretty the line.

• Overfitting: tuning rules until they match the past perfectly. Example: test 100 random rules and, at a usual 5% significance level, about 5 will look 'significant' by pure luck.
• Survivorship bias: testing only companies that still exist today. Example: testing on today's 500 largest companies over 20 years leaves out those that went bankrupt, so the past looks better than it was.
• Lookahead bias: using information you would not have had at the time. Example: deciding to buy at the open using that day's closing price.
• Ignoring costs and taxes: Example: a rule that trades 50 times a year at 0.2% per round trip loses about 10% a year to costs before any tax, and gains held under a year are often taxed at higher rates.

• I can explain why the rule should work, not only that it did work.
• I tested it on data it was not built on, and it still held up.
• I counted every version I tried, not just the best one.
• I included realistic costs, taxes and delays.
• I know the worst drawdown in the test and I can live with a worse one.
• I have an emergency fund, and this money is money I can afford to lose.
• I will follow the rules in a bad year without overriding them.
• I know a broad index fund would beat my strategy if it fails, and I accept that.

If you could not tick most of these, the index fund is the honest answer. Nothing here promises a return. Past results, whether a fund's or your own backtest's, do not guarantee future ones, and this course is education, not advice.