P2.T7.604. Model error and model implementation risk (Crouhy, Galai & Mark)

Nicole Seaman

Director of CFA & FRM Operations
Staff member
Subscriber
Learning objectives: Identify and explain errors in modeling assumptions that can introduce model risk. Explain how model risk can arise in the implementation of a model.

Questions:

604.1. According to Crouhy, Galai & Mark, JPMorgan's London Whale incident "showed that model risk has no respect for the size or standing of an institution." According to the US Senate's Subcommittee Report, despite portraying itself as an expert in risk management, the bank's Chief Investment Office (CIO) which was charted with managing excess deposits "placed a massive bet on a complex set of synthetic credit derivatives that, in 2012, lost at least $6.2 billion."

Which of the following is TRUE as risk management failure that contributed to the loss at JPMorgan's CIO?

a. Soon after breaching the bank' and CIO's VaR limit, a new VaR model was adopted (and approved) that reduced the SCP VaR by 50%, enabling the CIO to end its breach
b. The CIO switched from its historical practice of marking credit derivative positions at or near the midpoint price in the daily range to assigning the favorable price within the daily price range
c. SCP trades routinely breached the limits on all five key metrics used by CIO (ie, VaR, CS01, CSW10%, stress loss, and stop loss), and the breaches were reported to management, but the breaches were largely ignored
d. All of the above are true, according to the Senate Subcommittee: the loss was caused by failures in operational risk, model risk, and corporate governance


604.2. Crouhy, Galai & Mark explain that the main cause of model risk are either (i) model error or (ii) implementation. Model error is when "the model might contain mathematical errors or, more likely, be based on simplifying assumptions that are misleading or inappropriate." Implementation is when "the model might be implemented wrongly, either by accident or as part of a deliberate fraud. "

Each of the following is a classic example of how model error can be introduced EXCEPT which is the LEAST likely assumption, by itself, to create model error risk?

a. To assume an asset's distribution is stationary over time in order to maintain or improve the tractability of the model
b. To assume a delta-neutral hedging strategy is risk-free and can be maintained because active re-balancing is unrealistic
c. To assume asset returns follow an empirical distribution simply because the historical data happens to be easily available
d. To assume the forward rates--i.e., that are used to value fixed-income instruments--are log normal although interest rates have shifted into a long-term regime of negative territory


604.3. In regard to model implementation, Crouhy says "even if a model is correct and is being used to tackle an appropriate problem, there remains the danger that it will be wrongly implemented. With complicated models that require extensive programming, there is always a chance that a programming “bug” may affect the output of the model. Some implementations rely on numerical techniques that exhibit inherent approximation errors and limited ranges of validity. Many programs that seem error-free have been tested only under normal conditions and so may be error-prone in extreme cases and conditions. "

With respect to model implementation, which of the following is the BEST pieces of advice?

a. Ensure responsibility for data accuracy is clearly assigned
b. Remove outliers in all cases because outliers distort skew and kurtosis of the distribution
c. Volatility and correlation should be directly observed rather than forecast; if these two inputs cannot be observed, seek an alternative approach
d. Seek the maximum length of the sampling period in order to improve the power of statistical tests and reduce estimation errors

Answers here
 
Top