FRM二级考试即将到来，考生们临考前要回顾错题，查漏补缺。高顿网校FRM小编预祝大家获得好成绩！
　　1.Operational risk loss data is not easy to collect within an institution, especially for extreme loss data. Therefore, financial institutions usually attempt to obtain extemal data, but doing so may create biases in estimating loss distribution. Which of the following statements regarding characteristics of extemal loss data is incorrect?
　　A. Extemalloss data often exhibits scale bias as operational risk losses tend to be positively related to the size of the institution (i.e., scale ofits operations).
　　B. Extemalloss data often exhibits truncation bias as minimum loss thresholds for collecting loss data are not uniform across all institutions.
　　C. Extemalloss data often exhibits data capture bias as the likelihood that an operational risk loss is reported is positively related to the size ofthe loss.
　　D. The biases associated with extemalloss data are more important for large losses in relation to a bank's assets or revenue than for small losses.
　　2.When selecting between credit models, which of the following factors is least important?
　　A. That the model's parameter estimates are linear.
　　B. How easy the models are to understand.
　　C. How robust the models are when new data are added into the analysis.
　　D. The time to calibrate and recalibrate the model.
　　3.Which if the following statements best describes the process of "convolution" in the context of a specific application?
　　A. Systematically recording the probability that a random variable is less than or equal to each possible value.
　　B. Linking causes and effects through conditional probabilities.
　　C. Taking a time series of earnings and computing its volatility.
　　D. Combining a frequency and a severity distribution into an aggregate loss distribution.
　　Answer:
　　1.D
　　The biases associated with externalloss data are important for alllosses in relation to a bank's assets or revenue.
　　2.A
　　It is important for models to be understandable, robust and able to be recalibrated. Models do not have to be linear. However, a researcher may wish to use an easily understood model, yet she may have to choose a model that is very complex and takes a long time to calibrate because it gives the
　　most accurate results.
　　3.D
　　Convolution involves combining a frequency distribution and a severity distribution into an aggregate loss distribution.