Practice Test 1

In one sentence, what fundamentally distinguishes quantitative trading from traditional discretionary trading?

• It only trades cryptocurrencies and other new asset classes
• It replaces intuition and news reaction with data, probability and systematic testing
• It guarantees a profit on every individual trade it places
• It avoids using any historical data when making decisions

What is the correct order of the five-stage quant research workflow?

• Deploy → Backtest → Build signals → Analyze patterns → Collect data
• Build signals → Collect data → Analyze patterns → Deploy → Backtest
• Collect data → Analyze patterns → Build signals → Backtest → Deploy & monitor
• Analyze patterns → Collect data → Deploy → Build signals → Backtest

"Alpha" in quantitative trading is best defined as:

• A statistical edge — a signal that is right more often than chance, producing returns beyond the benchmark
• The total amount of leverage applied to a portfolio
• A guarantee that a strategy will win on 100% of its trades
• The fee a broker charges per executed order

In an OHLCV bar, what does the "C" represent?

• The total contract count outstanding
• The cumulative volume across the session
• The commission paid on the bar
• The closing price of the period

The bid-ask spread is calculated as:

• Bid price minus ask price
• Ask price minus bid price
• The average of the day's high and low
• Closing price minus opening price

Why do quants prefer to analyze returns rather than raw prices?

• Returns are always positive, which simplifies the maths
• Prices cannot be stored in a database but returns can
• Returns normalize price moves so assets at different price levels are comparable
• Returns remove the need to record any timestamps

Volatility, in the context of a return series, measures:

• The size of the swings (standard deviation of returns), not their direction
• The guaranteed direction the price will move next
• The average trading volume over the window
• The number of trades executed per day

Compared with a 50-day moving average, a 20-day moving average generally:

• Is smoother and lags more
• Ignores the most recent prices
• Can only be used on weekly data
• Reacts faster but stays noisier

The classic momentum (trend-following) trading rule using two moving averages is:

• BUY when the short MA falls below the long MA
• BUY when the short MA rises above the long MA, otherwise SELL
• BUY only when both MAs are perfectly flat
• SELL whenever price equals the long MA exactly

Which bias is widely regarded as the most dangerous in backtesting?

• Rounding error in price data
• Using too many decimal places
• Look-ahead bias — using information that would not have been available at decision time
• Storing data in the wrong file format

Survivorship bias occurs when a study:

• Includes only the assets or funds that survived, ignoring those that failed or were delisted
• Uses too small a sample of price ticks
• Annualizes volatility with the wrong constant
• Mislabels buy orders as sell orders

A market order differs from a limit order in that a market order:

• Always rests passively in the order book
• Can only be used by institutional traders
• Sets a price ceiling and never executes above it
• Prioritizes immediate execution by crossing the spread

"Liquidity" in a market refers to:

• How quickly a company pays its dividends
• The ability to trade size without significantly moving the price
• The total cash a broker holds in reserve
• The number of decimal places used in quotes

A mean-reversion strategy is built on the assumption that:

• Strong trends always continue in the same direction
• Volume always precedes price
• Extreme moves are temporary and prices tend to snap back toward a normal level
• Prices follow a perfect random walk with no anchor

A z-score of 0 for a price means the price is:

• Exactly at its mean (average) level
• Two standard deviations above the mean
• At an extreme low and a strong buy
• Impossible to compute

Why is pure (textbook) arbitrage considered risk-free while statistical arbitrage is not?

• Pure arbitrage is illegal, so it carries no market risk
• Statistical arbitrage uses no data at all
• Pure arbitrage always loses money slowly
• Pure arbitrage captures an exact simultaneous price gap, while stat arb bets on a pattern that can fail

For pairs trading, the property that actually guarantees the spread keeps returning to its mean is:

• High correlation between the two prices
• Cointegration — a stable long-term equilibrium relationship
• Both stocks being in the same index
• Both stocks having the same share price

The Sharpe ratio measures:

• The maximum peak-to-trough loss of a strategy
• The fraction of trades that are winners
• Excess return earned per unit of risk (volatility)
• The total number of trades placed per year

Maximum drawdown describes:

• The largest peak-to-trough decline in portfolio value
• The average daily return of the strategy
• The number of consecutive winning trades
• The spread paid on each round-trip trade

Diversification reduces risk most effectively when the assets are:

• Perfectly positively correlated (corr = 1)
• All the same single stock
• Identical in every respect
• Weakly or negatively correlated with each other

In machine learning for trading, "overfitting" means the model:

• Performs poorly on both training and test data
• Memorizes noise — high accuracy in training but poor out-of-sample results
• Uses too few parameters to learn anything
• Refuses to make any predictions

Which is a hallmark of a good feature for an ML trading model?

• It is purely random with no relationship to returns
• It secretly contains future information
• It is relevant, stable, and intuitive
• It changes meaning unpredictably every day

Walk-forward validation is preferred over a single train/test split because it:

• Repeatedly retrains and tests on rolling windows, giving a more realistic estimate
• Uses the entire dataset for training and never tests
• Eliminates the need to collect any data
• Guarantees the strategy will be profitable live

Why can ordinary tools like Excel struggle with raw market tick data?

• Tick data contains no timestamps
• Tick data is always smaller than OHLC data
• Excel cannot perform any arithmetic
• Markets emit massive volumes of every trade, quote and book update — far beyond Excel's capacity

In kdb+/q, the columnar storage model is fast for analytics mainly because it:

• Stores each row as a separate file on disk
• Reads only the columns a query actually needs
• Compresses all data into a single integer
• Forbids any time-based queries

Signal decay refers to the phenomenon where:

• A signal's code stops compiling over time
• The data feed is delayed by a few milliseconds
• An alpha weakens as markets adapt and competitors trade the same edge
• Volatility always rises after a signal fires

A spike in volume relative to its recent average is most directly used as a:

• Volume-based signal indicating unusual activity or conviction behind a move
• Guaranteed sell signal regardless of price
• Measure of the risk-free interest rate
• Replacement for the closing price

Why is a high win rate alone a misleading measure of a strategy's quality?

• Win rate cannot be calculated from historical data
• A high win rate always means high Sharpe
• Win rate is only defined for losing strategies
• A strategy can win often but still lose money if its few losses are very large

"Good backtest ≠ good strategy" is a central lesson because:

• Backtests are always negative
• Biases, overfitting and regime changes can make impressive historical results fail to repeat live
• Live trading uses different arithmetic than backtests
• Backtests cannot include transaction costs at all

The single best summary of what quant trading requires is:

• Luck + leverage + speed
• News + intuition + experience
• Data + research + probability + discipline
• A single perfect signal that never decays

A researcher observes that stocks which rose the most over the past 3 months continue to outperform over the next month. Which signal family does this describe?

• Mean reversion
• Momentum (trend continuation)
• Pairs / spread reversion
• Pure arbitrage

An analyst computes a feature as today's return using tomorrow's closing price by mistake. In the live system this feature would:

• Improve real performance because it is more accurate
• Have no effect since prices are random
• Be unusable live — it relies on future data (look-ahead / data leakage), inflating the backtest only
• Reduce transaction costs

You want to turn raw price and volume data into predictive inputs. Which set is the best example of feature engineering?

• Returns, rolling volatility, and the volume ratio $V_t/\text{avg}(V)$
• The company's CEO name and ticker symbol
• The raw closing price copied without transformation
• The exchange's street address

Two signals have identical average returns, but Signal X's edge is stable across every year tested while Signal Y's edge appears only in one year. Which should you trust more, and why?

• Signal Y, because concentrated returns are always better
• Neither — average return is the only thing that matters
• Signal Y, because it is newer
• Signal X, because signal stability across periods indicates a more robust, repeatable edge

A pairs trader watches the spread between Coke and Pepsi. The spread's z-score hits +2.5. The correct market-neutral action is:

• Buy both Coke and Pepsi outright
• Short the spread: short the richer leg, long the cheaper leg, expecting convergence
• Long the spread: buy the richer leg, short the cheaper leg
• Do nothing — z-scores are irrelevant to pairs

A backtest shows a 200% annual return with an unrealistically smooth equity curve and uses 47 tuned parameters. The most likely diagnosis is:

• A genuinely excellent strategy ready to deploy
• A data feed outage
• Overfitting — too many parameters and over-optimization fitting historical noise
• Survivorship bias only, with no overfitting

A strategy's gross backtest return is strong, but after subtracting commissions and slippage it turns negative. The key lesson is:

• Transaction costs and slippage must be modelled; net return is what actually matters
• Gross return is the only figure that counts
• Slippage helps high-frequency strategies
• Costs can be ignored if the win rate is high

A quant has a hypothesis that high relative volume predicts next-day returns. Following the research pipeline, what is the correct next step after forming the hypothesis?

• Deploy capital live immediately to test it
• Optimize parameters until the result looks perfect
• Publicly announce the edge to competitors
• Gather data and engineer features, then build and backtest the signal before any deployment

A spread between two stocks has been highly correlated for years, but one company is acquired in a merger and the relationship snaps. This illustrates that:

• Correlation guarantees cointegration
• Pairs can break on structural change, so cointegration must be monitored continuously
• Mergers always make pairs more profitable
• Z-scores cannot be computed after a merger

A model scores $R^2 = 0.92$ on training data but $R^2 = -0.15$ on the test set. This pattern is the signature of:

• An underfit model that is too simple
• A perfectly generalizing model
• Overfitting — the model memorized training noise and fails to generalize
• A data feed with no observations

An equal-weight portfolio of 5 strategies (momentum, mean reversion, breakout, pairs, carry) is built. The main benefit of combining diverse strategy types is:

• Diversification across uncorrelated ideas smooths returns and reduces reliance on any single edge
• It guarantees each strategy individually never loses
• It removes the need to manage risk
• It eliminates all transaction costs

During the COVID crash, a strategy tuned on calm bull-market data suddenly failed. The best explanation is:

• The code had a syntax error
• The strategy's win rate was too high
• Transaction costs fell to zero
• A market regime change — the conditions the strategy relied on no longer held

Two strategies both return 10% per year. Strategy P has volatility 5%; Strategy Q has volatility 20%. On a risk-adjusted basis:

• Q is better because higher volatility means higher returns
• P is better — same return for far less risk gives a higher Sharpe ratio
• They are identical because returns are equal
• Neither can be compared without the share price

A researcher only studies the 30 stocks currently in an index to "prove" a strategy worked over 20 years. The flaw is:

• Look-ahead bias from lagging the signal
• Annualizing volatility incorrectly
• Survivorship bias — today's index members exclude the companies that failed or were removed
• Using too many parameters

A signal's Information Coefficient (IC) — the correlation between its predictions and realized returns — drops steadily month after month. This most likely indicates:

• Signal decay as the edge is competed away and the market adapts
• The signal is becoming stronger over time
• A data storage error that is harmless
• That IC and returns are unrelated by definition

A stock moves from \$80 to \$86 in one day. What is the simple return?

• 6.0%
• 8.0%
• 7.5%
• 7.0%

A stock moves from \$200 to \$210. What is the log return (to two decimals)?

• 5.00%
• 4.88%
• 5.13%
• 4.76%

Daily returns over three days are +20%, −10%, +5%. What is the cumulative return?

• 13.4%
• 15.0%
• 14.0%
• 12.5%

Prices are [40, 44, 42, 46, 48]. What is the latest 4-day simple moving average (SMA)?

• 44.0
• 46.0
• 43.0
• 45.0

Daily returns are [3%, −2%, 4%, −1%, 1%]. What is the mean (average) return?

• 0.5%
• 1.0%
• 2.0%
• 1.5%

Returns are [1%, −1%, 2%, −2%]. Using the sample standard deviation ($n-1$), the rolling volatility is closest to:

• 1.58%
• 2.00%
• 1.83%
• 1.50%

A strategy placed 45 trades and 27 were profitable. What is the hit rate?

• 45%
• 67%
• 54%
• 60%

Six trades returned [+5%, −3%, +8%, −2%, +4%, −2%]. What is the average return per trade (to two decimals)?

• 1.67%
• 2.00%
• 1.50%
• 3.33%

Today the 10-day (fast) SMA is 52 and the 30-day (slow) SMA is 50. Under the standard momentum crossover rule, the signal is:

• SELL, because the fast MA is above the slow MA
• BUY, because the fast MA is above the slow MA
• EXIT, because the two MAs are equal
• No signal — crossovers are never tradeable

A price is \$88. Its rolling mean is \$100 and rolling standard deviation is \$8. What is the z-score?

• +1.5
• −1.2
• −1.5
• +1.2

A mean-reversion model uses ±2 thresholds. The current z-score is −2.4. What action does the rule trigger?

• BUY — the price is oversold ($z < -2$), expect reversion upward
• SELL — the price is overbought
• EXIT — the price is at its mean
• Do nothing — −2.4 is within the band

Today's volume is 4.5 million shares; the 20-day average volume is 1.5 million. What is the volume ratio (spike factor)?

• 1.5×
• 0.33×
• 2.0×
• 3.0×

A strategy returns 12% with a risk-free rate of 2% and volatility of 8%. What is the Sharpe ratio?

• 1.50
• 1.25
• 1.75
• 0.80

A portfolio peaks at \$120 and later falls to a trough of \$90. What is the maximum drawdown?

• −33.3%
• −30.0%
• −25.0%
• −20.0%

A strategy's gross return is 8.0%. Transaction costs are 1.2% and slippage is 0.3%. What is the net return?

• 6.5%
• 6.8%
• 7.7%
• 5.3%

Three trades execute at (price, volume) of (\$10, 100), (\$12, 300) and (\$11, 100). What is the VWAP?

• 11.0
• 12.0
• 11.2
• 11.4

A signal wins 60% of the time. The average win is \$150 and the average loss is \$100. What is the expected value per trade?

• \$30
• \$50
• \$25
• \$90

Over a holding period a position's price rises from \$50 to \$60. What is the total (cumulative) return?

• 16.7%
• 10.0%
• 20.0%
• 25.0%

Daily volatility is 2%. Using the square-root-of-time rule with 252 trading days, the annualized volatility is closest to:

• 504%
• 2.0%
• 50.4%
• 31.7%

A cointegrated pair's spread has mean 10 and standard deviation 2. The current spread is 14. What is the spread's z-score and the correct action?

• $z = +2$ → SHORT the spread (short A, long B), expecting reversion
• $z = +2$ → LONG the spread (long A, short B)
• $z = -2$ → BUY the spread
• $z = +4$ → do nothing