引導式測驗 · 中等 等級

機器學習因子投資測驗

測試因子定義、時間感知 ML 驗證,以及為什麼預測收益在扣除投資組合成本後可能無法產生可實現的回報。

相關引導路徑機器學習因子投資
問題 1 的 8

What does a factor strategy use to rank or group securities?

預覽每個問題8 問題

當您參加測驗時,就會出現答案和解釋。

  1. What does a factor strategy use to rank or group securities?
    • A model architecture independent of its input data
    • A trading venue’s order-matching priority
    • A defined characteristic or portfolio signal, such as value or momentum
    • A guarantee that every selected security will earn excess returns
  2. What can ML add to a traditional factor score?
    • It can estimate nonlinear relationships or interactions among characteristics
    • It removes the need to choose a target or benchmark
    • It makes historical data independent across time
    • It turns a backtest into a live performance record
  3. Why use chronological train and test periods in a return-prediction study?
    • To maximize the number of stocks in each training batch
    • To ensure that all factor exposures stay constant
    • To eliminate transaction costs from the evaluation
    • To avoid using later market information when fitting predictions for earlier dates
  4. What does a higher out-of-sample prediction score establish by itself?
    • That the signal is causal rather than correlated
    • Better prediction under that test design, not guaranteed excess portfolio returns
    • A durable factor premium in every future market
    • Net profitability after turnover and market impact
  5. Why should a factor backtest account for turnover and market impact?
    • Market impact matters only for equally weighted benchmarks
    • Transaction costs are already captured by an in-sample prediction score
    • Trading costs can reduce or erase returns from frequently changing signals
    • Turnover determines whether a characteristic is a factor by definition
  6. How should a team interpret an ML feature-importance chart when predictors are correlated?
    • As a guarantee that the most important feature will persist
    • As a descriptive and potentially unstable attribution, not proof that one feature causes returns
    • As a causal decomposition that is invariant to the model
    • As evidence that omitted features have no influence
  7. Which report gives a reviewer enough context to assess a factor result?
    • The signal definition, benchmark, sample, portfolio construction, turnover, cost assumptions, and holdout design
    • Only the largest historical annual return
    • A model name and one in-sample accuracy value
    • A chart without dates or universe definitions
  8. How should an analyst frame a historical factor premium?
    • It guarantees that any fund using the factor will outperform
    • It proves the factor will work in every country and market regime
    • It makes benchmark and cost choices irrelevant
    • It is an empirical result under a particular definition, period, and implementation, not a promise of future returns