▸case-01 We are preparing a daily equity price dataset for quantitative backtesting. There are missing values on non-trading days and random missing ticks during trading hours. A junior analyst suggested filling missing prices with the global column mean to preserve overall distribution stats. Provide the recommended fill strategy for time series price data. | pass→pass | 18,217 | 44,421 | +144% | 1 | 1 | 0% | 2,894 | 2,416 | -17% | 0 | 0 | — |
▸case-02 When analyzing daily log returns for a cryptocurrency dataset, the data exhibits heavy tails and high skewness. Should we rely on standard Z-Score with a threshold of 3.0 to strip outliers, or an alternative metric resilient to non-normal distributions? Specify the appropriate statistical method. | pass→pass | 15,759 | 14,238 | -10% | 1 | 1 | 0% | 2,650 | 2,531 | -4% | 0 | 0 | — |
▸case-03 We have tick data from Binance, Coinbase, and Kraken recorded with slightly unsynchronized timestamps and differing sampling frequencies. We need to construct a unified pandas DataFrame across all three exchanges for cross-sectional momentum. Describe the required data alignment approach. | pass→pass | 23,324 | 20,207 | -13% | 1 | 1 | 0% | 3,825 | 3,657 | -4% | 0 | 0 | — |
▸case-04 In a multi-factor equity dataset containing P/E, P/B, EV/EBITDA, and Debt-to-Equity, some individual factors look plausible individually but represent impossible combinations jointly. Standard univariate bounds miss these points. What detection technique handles multi-dimensional anomalies? | pass→pass | 16,669 | 16,340 | -2% | 1 | 1 | 0% | 2,725 | 2,637 | -3% | 0 | 0 | — |
▸case-05 We have cross-sectional factor exposure values that have been transformed and verified to follow a near-Gaussian distribution across S&P 500 stocks. We need a simple, parametric metric to flag observations beyond 3 standard deviations. What method should be applied? | pass→pass | 9,404 | 8,433 | -10% | 1 | 1 | 0% | 1,559 | 1,677 | +8% | 0 | 0 | — |
▸case-06 A high-frequency exchange feed emitted repeated timestamp records with slightly updated bid/ask quotes due to network re-transmissions. We need to clean the pandas DataFrame to remove duplicate timestamps. Should we drop all duplicates, keep the first, or keep the last quote received? | pass→pass | 12,349 | 9,654 | -22% | 1 | 1 | 0% | 1,937 | 1,664 | -14% | 0 | 0 | — |
▸case-07 We are feeding bond yields into a neural network that strictly requires feature values compressed between 0 and 1. Standardizing to zero mean and unit variance leaves values outside [0, 1]. Which scaling technique explicitly bounds features into a fixed [0, 1] interval? | pass→pass | 8,538 | 8,407 | -2% | 1 | 1 | 0% | 1,438 | 1,577 | +10% | 0 | 0 | — |
▸case-08 Before running an OLS risk model regression, factors have drastically different units (e.g., market cap in billions vs dividend yield in percentages). The model requires centered features with unit standard deviation. What data transformation should be performed? | pass→pass | 6,992 | 10,676 | +53% | 1 | 1 | 0% | 1,193 | 2,085 | +75% | 0 | 0 | — |
▸case-09 An intraday 1-minute high-frequency dataset has isolated 2-minute gaps where trades were scarce, but market liquidity remained continuous. Rather than holding stale prices with forward fill, we want smooth transition estimates between known endpoints. Which method estimates intermediate points continuous along a straight line? | pass→pass | 5,435 | 7,100 | +31% | 1 | 1 | 0% | 937 | 1,331 | +42% | 0 | 0 | — |
▸case-10 A corporate fundamental dataset contains multiple quarterly earnings filings for the same fiscal quarter due to amended 10-Q disclosures over time. How should duplicate quarter records be cleaned to reflect the most accurate historical picture for backtesting? | pass→pass | 15,950 | 15,383 | -4% | 1 | 1 | 0% | 2,600 | 2,835 | +9% | 0 | 0 | — |
▸case-11 An extreme outlier spike in raw trading volume threatens to skew linear factor loadings. Rather than completely dropping the row and losing valid price data, how should extreme values beyond 1.5 times the IQR be handled during cleaning? | pass→pass | 15,057 | 14,926 | -1% | 1 | 1 | 0% | 2,406 | 2,686 | +12% | 0 | 0 | — |
▸case-12 We have irregular tick data collected over multiple global timezones and need to generate uniform 24-hour daily trading bars ending at 16:00 EST. What cleaning/transformation step formats irregular times into fixed interval bars? | pass→pass | 14,010 | 14,783 | +6% | 1 | 1 | 0% | 2,521 | 2,683 | +6% | 0 | 0 | — |
▸case-13 In a cross-sectional snapshot of 1000 stocks at a single point in time, 5 stocks are missing market cap data. The sector distribution is verified uniform and stationary across the universe. What static central tendency metric imputes missing values based on average cross-sectional level? | fail→pass | 11,625 | 11,708 | +1% | 1 | 1 | 0% | 1,802 | 2,071 | +15% | 0 | 0 | — |
▸case-14 A security master database imported symbol strings containing trailing spaces and inconsistent letter casing (e.g., 'AAPL ', 'aapl', 'AAPL'). What initial string cleaning operations resolve these duplicate entities? | pass→pass | 9,530 | 8,662 | -9% | 1 | 1 | 0% | 1,597 | 1,717 | +8% | 0 | 0 | — |
▸case-15 For an asset return series with first quartile Q1 = 0.01 and third quartile Q3 = 0.05, calculate the upper fence cutoff for anomaly detection using the standard 1.5 multiplier. | pass→pass | 4,032 | 3,366 | -17% | 1 | 1 | 0% | 803 | 744 | -7% | 0 | 0 | — |
▸case-16 A historical macroeconomic database has a column with 95% missing entries across 20 years. An analyst wants to use linear interpolation to fill all missing values. What data cleaning rule should apply instead when missingness exceeds severe thresholds? | pass→pass | 11,051 | 14,044 | +27% | 1 | 1 | 0% | 1,676 | 2,207 | +32% | 0 | 0 | — |
▸case-17 We need to align macro indicators updated monthly on irregular announcement dates with daily equity prices without introducing look-ahead bias. What pandas merging or alignment method matches daily records to the most recent historical macro publication? | pass→pass | 12,521 | 11,970 | -4% | 1 | 1 | 0% | 2,296 | 2,299 | +0% | 0 | 0 | — |
▸case-18 We need to scale factor features for a quantitative machine learning pipeline. The features contain severe outliers that would shrink valid non-outlier data if Min-Max scaled or standard scaled. What scaler uses median and interquartile range? | pass→pass | 6,239 | 6,662 | +7% | 1 | 1 | 0% | 1,004 | 1,303 | +30% | 0 | 0 | — |
▸case-19 During initial dataset cleaning, we observe that a feature column 'trading_status_code' has the exact same scalar value '1' across all 500,000 historical rows. How should this feature be handled prior to model training? | pass→pass | 10,394 | 11,327 | +9% | 1 | 1 | 0% | 1,605 | 1,847 | +15% | 0 | 0 | — |
▸case-20 We have finished cleaning our financial feature matrix. Now we want to optimize a LightGBM model. What learning rate, max depth, and num_leaves settings should we set to maximize AUC on our validation set? | pass→pass | 15,427 | 14,244 | -8% | 1 | 1 | 0% | 2,823 | 2,720 | -4% | 0 | 0 | — |
▸case-21 Using clean historical price and volume data, design a mathematical formula for a novel cross-sectional momentum alpha factor that predicts 5-day future equity returns. | pass→pass | 36,367 | 21,417 | -41% | 1 | 1 | 0% | 6,390 | 4,155 | -35% | 0 | 0 | — |
▸case-22 Given a clean matrix of expected stock returns and asset covariance, compute the Markowitz mean-variance optimal weights subject to a long-only zero-leverage constraint. | pass→pass | 15,596 | 19,811 | +27% | 1 | 1 | 0% | 3,112 | 4,208 | +35% | 0 | 0 | — |