Textbook Quiz – Week 2

Answer each question in your own words. Be specific.

Your full name *0 / 150

Your institutional email address *Must match the address on your course enrolment. Checked against the roster.

Your class *Used to keep each class cohort’s records separate.

Which review period is this? *Submit once per period. Earlier submissions are kept, not replaced.

1. A signal can be implemented at three frequencies: | Rebalancing frequency | Expected gross alpha per month | Expected trading cost per month | |—|—:|—:| | Daily | 120 bps | 90 bps | | Weekly | 95 bps | 35 bps | | Monthly | 50 bps | 10 bps | Ignoring other risks, which frequency maximizes expected net alpha? A) Daily B) Weekly C) Monthly D) All three produce the same net alpha. Type the letter (A, B, C, or D) of the best answer. *0 / 10

2. A vendor delivers a 30-gigabyte file containing multirow headers, inconsistent date formats, blanks, text values embedded in numeric columns, and changing field definitions. Which architecture is most defensible? A) Preserve raw file, hashes, schema, typed tables, and a codebook B) Open the file manually in a spreadsheet and overwrite any values that appear unusual. C) Retain only the final clean table and delete the raw file to avoid duplication. D) Convert all fields to strings so that type conflicts disappear. Type the letter (A, B, C, or D) of the best answer. *0 / 10

3. A stock closes at $150. Overnight, it undergoes a three-for-one split and pays a $3 special dividend measured on the pre-split share basis. Apart from these events, its economic value is unchanged. What is the approximate theoretical ex-action price? A) $50 B) $51 C) $147 D) $49 Type the letter (A, B, C, or D) of the best answer. *0 / 10

4. A quarterly accounting variable is unavailable for a company because its filing has not yet been released. A researcher forward-fills the previous quarter's value and labels the observation as current. What is the better treatment for a point-in-time model? A) Use the company's subsequently reported figure because it describes the same fiscal quarter. B) Replace the value with the contemporaneous cross-sectional maximum. C) Keep it missing until release and use only information known then D) Linearly interpolate between the prior and subsequently reported quarters. Type the letter (A, B, C, or D) of the best answer. *0 / 10

5. A simplified VECM equation is delta y_t = -0.30 (y_{t-1} – 1.2 x_{t-1}) + epsilon_t. At t-1, the long-run disequilibrium term (y_{t-1} – 1.2 x_{t-1}) equals +5. Ignoring the innovation and all other terms, what is the expected contribution to delta y_t? A) -1.5 B) +1.5 C) +6.0 D) -0.3 Type the letter (A, B, C, or D) of the best answer. *0 / 10

6. A macro series is revised months after first release. A backtest uses the final revised history as if it were known on each decision date. What is the central data error? A) It should drop all observations that were later revised. B) It reports too many observations for a macroeconomic variable. C) It should convert every release to a string before modeling. D) It uses information unavailable at the historical decision time. Type the letter (A, B, C, or D) of the best answer. *0 / 10

7. A credit model uses the agency ratings AAA, AA, A, BBB, BB, B. A researcher encodes them as 6,5,4,3,2,1 and uses the result as a linear predictor of default probability. This specification assumes that the economic distance from AA to A is identical to the distance from BB to B. Which redesign best preserves the ordering while avoiding the equal-distance assumption? A) Target-encode each rating using default outcomes from the entire dataset. B) Replace the rating with an issuer-identifier variable. C) Standardize the integers to have mean zero and variance one. D) Use cumulative thresholds or another monotone ordinal encoding Type the letter (A, B, C, or D) of the best answer. *0 / 10

8. A dataset contains 100,000 corporate bonds, of which 1,000 default during the forecast horizon. A classifier that always predicts "no default" is therefore 99% accurate. Which procedure is most appropriate? A) Randomly split the data and select the model with the highest raw accuracy. B) Use chronological splits, rebalance training only, and evaluate default recall C) Oversample defaults before splitting the data so that identical default observations appear in training and testing. D) Undersample surviving bonds equally in the training, validation, and test sets. Type the letter (A, B, C, or D) of the best answer. *0 / 10

9. A nonlinear model predicts bond excess returns using yield-curve slope, sector, leverage, and liquidity. The partial dependence plot for yield-curve slope is nearly flat. Individual conditional expectation plots, however, show a strongly positive slope for banks and a strongly negative slope for regulated utilities. What is the best interpretation? A) The individual conditional expectation plots must be erroneous because they disagree with the partial dependence plot. B) Investigate sector conditioning or a sector-by-slope interaction C) Yield-curve slope has no predictive value in any part of the sample. D) Sector should be removed so that one common slope can be estimated. Type the letter (A, B, C, or D) of the best answer. *0 / 10

10. A bond dataset has 80 standardized liquidity and trading features. Pairwise distances are highly concentrated, so many normal observations appear almost equally far apart. One suspicious bond is not extremely distant from the sample, but its neighboring observations all lie in approximately the same direction from it, producing a very narrow range of angles. Which method is most likely to identify this bond? A) Replacement of every feature with its cross-sectional median B) A rule that flags only the largest Euclidean distance C) Angle-based anomaly detection D) Univariate winsorization of each feature independently Type the letter (A, B, C, or D) of the best answer. *0 / 10

Confirmation *I confirm that this submission is accurate and may be used for course administration.