ECO00042 Topics in Financial Econometrics Exercise Set 2 (10%) Spring 2023
Hello, dear friend, you can consult us at any time if you have any questions, add WeChat: daixieit
ECO00042 Topics in Financial Econometrics
Exercise Set 2 (10%) Spring 2023
1. Consider the daily IBM log daily returns (Tsay) over July 1962 - Septem- ber 1997 with a total of 9190 observations. We use two volatility models to calculate VaR of 1-day horizon at r = 9190 for a long position of $10m.
(a) Case 1: Assume that εt is standard normal. The fitted model is
pt = 0.00066 - 0.0247pt _2 + at , at = 7t εt
7t(2) = 0.00000389 + 0.0799at(2)_ 1 + 0.90737t(2)_ 1
From the data we have p9188 = 0.0153, p9189 = -0.0021, p9190 = -0.0128 and 79(2)190 = 0.000349. Then, derive VaR at 5% and 1% quantiles. (5 marks)
(b) Case 2: Assume that εt is a t-distribution with 5 dof. The fitted
model is
pt = 0.0003 - 0.0335pt _2 + at , at = 7t εt
7t(2) = 0.000003 + 0.0559at(2)_ 1 + 0.9357t(2)_ 1
Then, derive VaR at 5% and 1% quantiles. (5 marks)
(c) Describe the RiskMetrics approach to forecasting the daily value- at-risk (VaR) at 1 and 10 day horizons. Then, briefly describe the potential weakness of this approach. (7 marks)
(d) Inconsistency of VaR (see Frey and McNeil, 2002): Consider a port- folio of m = 50 defaultable corporate bonds. Assume defaults are independent, with common probability 2%. Face value is 100 which is repaid at T = r + ∆r if there is no default; otherwise there is no payment. The current price (T = r) is 95. The loss for the ith bond is the random variable given by
Li = -(100(1 - yi ) - 95) = 100yi - 5
where yi = 1 if default occurs and 0 otherwise. Then, Li ’s are a sequence of iid random variables with Pr(Li = -5) = 0.98 and Pr(Li = 95) = 0.02. Portfolio A is fully concentrated and comprises 100 units of Bond 1 with total price 9500. Portfolio B is fully diver- sified and comprises 2 units of each of the 50 bonds, again with total price 9500. Show that VaR is not a coherent risk measure in this example. (6 marks)
(e) Develop an alternative consistent measure which satisfies the subad- ditivity condition. (6 marks)
2. Agent believes that asset A will rise more than rest of market, and goes long (buys one unit) at time r - 1 at (log) price 夕A,t _ 1 . To protect against a fall, agent sells b units of asset B short at time r - 1 at price 夕B,t _ 1 for delivery at time, r. To satisfy the short sale contract the agent must buy βt _ 1 units of asset B at price, 夕B,t . Then, an agent gets a return equal to
∆夕A,t - βt _ 1 ∆夕B,t
(a) Derive the time-varying optimal hedge ratio. (4 marks)
(b) Describe four modelling approached and discuss their empirical per- formance in terms of the hedging variance reduction and the mini- mum capital requirement associated with VaR. (8 marks)
(c) We now wish to combine the bivariate error correction model with the bivariate DCC-GARCH model. Derive this joint modelling approach in detail, and discuss why this approach is likely to improve the empirical performance. (8 marks)
3. Consider the following panel data model:
yit = βzit + 7zi + εit , i = 1, ..., N; r = 1, ..., T,
εit = ai + tit
where yit is a scalar dependent variable, zit a scalar time-varying strictly exogenous regressor, zi is a scalar time-invariant regressor, ai ’s are indi- vidual specific effects and tit ’s are idiosyncratic disturbances distributed independently over both cross-sections and time periods with zero mean and finite variance 7u(2) . Here we assume that ai ’s are distributed with zero mean and finite variance 7α(2), but also distributed independently of tit and that N is sufficiently large and T is fixed.
(a) Assume that zit and zi are both uncorrelated with ai . Then, show that both the OLS estimator and the random effects estimators of β and 7 are consistent but the random effects estimator is more efficient. (9 marks)
(b) We now assume that ai ’s are correlated with zit and zi . What hap- pens to the OLS and the random effects estimators of β and 7? (7 marks)
(c) Decompose zit = (z1,it, z2,it) and zi = (z1i, z2i). Assume that ai ’s are uncorrelated with z1,it and z1i, but correlated with z2,it and z2i . Then, show that the Fixed Effects estimator provides consistent es- timator of β’s only. Then, derive the Hausman-Taylor IV estimation techniques for consistently estimating 71 and 72 . (9 marks)
4. Short essay questions:
(a) The empirical performance of parametric, nonparametric and semi- parametric VaR models in terms of Backtestings. (13 marks)
(b) Suppose that there exists cross-section dependence in εit as follows: εit = ai + e9t + tit
where 9t is the p × 1 vector of unobserved (time varying) factors, which may be correlated with zit . Describe the two main approaches for consistently estimating β . Then, develop the extended Hausman- Taylor IV estimation for consistently estimating V . (13 marks)
Notes: The second in-course exercise (10%) will be downloadable from VLE on Friday 24 Feb., and one anonymous copy of this assignment only with student number and exam number must be submitted to the online Economics Submission Point no later than 3PM on Monday 6 March. Handwritten reports are accepted. Failure to meet this initial deadline will result in a reduction of marks (please see VLE for details)
2023-05-08