NO.PZ202110140100000408
问题如下:
Emily Yuen is a senior analyst for a consulting firm that specializes in assessing
equity strategies using backtesting and simulation techniques. She is working with an
assistant, Cameron Ruckey, to develop multifactor portfolio strategies based on nine
factors common to the growth style of investing. To do so, Yuen and Ruckey plan to
construct nine separate factor portfolios and then use them to create factor-weighted
allocation portfolios.
Yuen tasks Ruckey with specifying the investment universe and determining the
availability of appropriate reporting data in vendor databases. Ruckey selects a vendor
database that does not provide point-in-time data, so he adjusts the database to include
point-in-time constituent stocks and a reporting lag of four months.
Next, Yuen and Ruckey run initial backtests on the nine factor portfolios, calculating performance statistics and key metrics for each. For backtesting purposes, the portfolios are rebalanced monthly over a 30-year time horizon using a rolling-window procedure. Yuen and Ruckey consider a variety of metrics to assess the results of the factor portfolio backtests. Yuen asks Ruckey what can be concluded from the data for three of the factor strategies in Exhibit 1:
Exhibit 1 Backtest Metrics for Factor Strategies
Ruckey tells Yuen the following:
Statement 1 We do not need to consider maximum drawdown, because standard deviation sufficiently characterizes risk.
Statement 2 Factor 2 has the highest downside risk.
From her professional experience Yuen knows that benchmark and risk parity
factor portfolios, in which factors are equally weighted and equally risk weighted,
respectively, are popular with institutional and high-net-worth clients. To gain a more
complete picture of these investment strategies’ performance, Yuen and Ruckey design
a Benchmark Portfolio (A) and a Risk Parity Portfolio (B), and then run two simulation
methods to generate investment performance data based on the underlying factor
portfolios, assuming 1,000 simulation trials for each approach:
Approach 1 Historical simulation
Approach 2 Monte Carlo simulation
Yuen and Ruckey discuss the differences between the two approaches and then
design the simulations, making key decisions at various steps. During the process,
Yuen expresses a number of concerns:
Concern 1: Returns from six of the nine factors are correlated.
Concern 2: The distribution of Factor 1 returns exhibits excess kurtosis and
negative skewness.
Concern 3: The number of simulations needed for Approach 1 is larger than the
size of the historical dataset.
For each approach, Yuen and Ruckey run 1,000 trials to obtain 1,000 returns for Portfolios A and B. To help understand the effect of the skewness and excess kurtosis observed in the Factor 1 returns on the performance of Portfolios A and B, Ruckey suggests simulating an additional 1,000 factor returns using a multivariate skewed Student’s t-distribution, then repeating the Approach 2 simulation.
The process Ruckey suggests to better understand how the performance of Portfolios A and B using Approach 2 is affected by the distribution of Factor 1 returns is best described as:
选项:
A.data snooping.
B.sensitivity analysis.
C.inverse transformation.
解释:
B is correct.
Sensitivity analysis can be implemented to help managers understand how the target variable (portfolio returns) and risk profiles are affected by changes in input variables. Approach 2 is a Monte Carlo simulation, and the results depend on whether the multivariate normal distribution is the correct functional form or a reasonable proxy for the true distribution. Because this information is almost never known, sensitivity analysis using a multivariate skewed Student’s t-distribution helps to account for empirical properties such as the skewness and the excess kurtosis observed in the underlying factor return data.
A is incorrect. Data snooping is the subconscious or conscious manipulation of
data in a way that produces a statistically significant result (i.e., a p-value that is
sufficiently small or a t-statistic that is sufficiently large to indicate statistically
significance).
C is incorrect. The inverse transformation method is the process of converting a randomly generated number into a simulated value of a random variable.
不太明白為何A不對?