Model validation task.
Introduction.
Assume that you work for a company that trades European Equity Options. Assume that your company’s
IT system uses the Black-Scholes formula for pricing these options and the dividends are modeled as a
continuous yield.
Let’s denote by S the today’s equity spot price, r – the interest rate, q – dividend yield, K – option strike,
T – time to maturity (in years) and σ – the option volatility. In the Black-Scholes model a Call option price
is calculated as
𝐶 𝐾, 𝑇 = 𝑒 −𝑟𝑇 [𝐹𝑁 𝑑1 − 𝐾𝑁(𝑑2 )]
, where
𝑟−𝑞 𝑇
𝐹 = 𝑆𝑒 ,
1 𝐹 𝜎2
𝑑1 = [ln + 𝑇],
𝜎 𝑇 𝐾 2
𝑑2 = 𝑑1 − 𝜎 𝑇,
N(.) - standard (0,1)-Gaussian cumulative distribution function.
A Put option price is calculated similarly:
𝑃 𝐶, 𝐾 = 𝑒 −𝑟𝑇 [𝐾𝑁 −𝑑2 − 𝐹𝑁(−𝑑1 )]
You will be asked to do two tasks similar to those performed during validation of a model used for
calculating Value-at-Risk (VaR) arising from moves in equity spot prices (will be denoted by EQVAR
below). EQVAR is defined as the amount of cash such that the change of a portfolio’s price due to equity
spot price moves from one date to the next will be above it with probability 99%. In other words, EQVAR
is the 1% percentile of potfolio price changes arising from equity price changes.
EQVAR calculation requires a model, please use the one described below. The model uses the historical
simulation approach. We assume that t=0 corresponds to "today", t=1 corresponds to "tomorrow" and
negative values of t correspond to values in the past. Also, by N we denote the number of equity
underlyings the portfolio price depends on.
The model proceeds as follows:
1. For every equity underlying i obtain 3 years of historical prices. Use this data to calculate 3 years
𝑆𝑡+1,𝑖 −𝑆𝑡,𝑖
(t=-750..-1, i.e. 750 business days) of historical equity returns 𝑟𝑡,𝑖 = .
𝑆𝑡,𝑖
2. For every underlying i and for each historical date t apply the corresponding return to today’s
(𝑡)
price to obtain a hypothetical tomorrow's equity price 𝑆1,𝑖 = 𝑆0,𝑖 (𝑟𝑡,𝑖 + 1). Thus we obtain 750
sets of tomorrow's underlying prices. Calculate the portfolio price with these new, modified
equity prices. Thus way we obtain ~750 portfolio price changes
𝑡 𝑡 𝑡
𝑃𝑁𝐿𝑡 = 𝑉(𝑆1,1 , 𝑆1,2 , . . . , 𝑆1,𝑁 ) − 𝑉(𝑆0,1 , 𝑆0,2 , . . . , 𝑆0,𝑁 )
, where V() is the portfolio price.
3. We assume that PnLt are samples of a random variable distribution. The VaR is
calculated as the 1% interpolated percentile of those 750 samples.
The task
You are provided time-series of 3 equity underlyings for this task: Bank of America, Microsoft, Apple.
1. The Historical Simulation approach relies on the assumption that the distribution of returns is
stationary, i.e. does not change very fast. Think of a way to test this assumption and run this
test.
2. As a high-level check of the methodology you are required to run backtesting: see how many
times the actuall losses would exceed VaR during the last year (i.e. the drop in portfolio price
was more significant than VaR). For this test assume that for all dates in the past r=0.03, and for
each underlying σ=0.2, q=0.01. In other words, for this task assume that all pricing parameters
except the equity spot price are fixed.
The portfolio of the company is made up of the following holdings:
100 Call options on Bank of America with maturity=2Mar2015 and K=16
30 Put options on Microsoft with maturity=2Mar2015 and K=40
3 Call options on Apple with maturity=2Mar2015 and K=600
The company’s portfolio has been without changes during the past year.
As result of this task you should produce an Excel (or LibreOffice) file where you do the analysis and a
text document describing your thinking and conclusions (in English). If you choose, you may use another
system (not Excel), the only requirement is that you must provide all the files where the testing was
performed. In other words we would like to be able to reproduce your test results.