CITY UNIVERSITY OF HONG KONG Course code & title : MS6217 Statistical Modelling in Finance Session : Semester B, 2004-2005 Time allowed : 3 hours This paper has 13 pages (including this page) Instructions to candidates: 1. 2. Answer ALL FOUR questions Show sufficient work for each question Materials, aids and instruments permitted during examination: Approved calculator 1 Question 1 (25 marks) Consider the following transfer function model, Yt v( B) X t N t (1.1) where v( B) vo v1 B v 2 B 2 ....... , and B is the backward shift operator. a) An equivalent way of representing (1.1) is Yt w( B) X t b / ( B) q ( B) Q ( B L )at / p ( B) P ( B L ) . (1.2) Explain the purposes of w(B), (B) and b in (1.2). Explain also why (1.2) is an equivalent way of representing (1.1). (5 marks) b) To examine the relationship between the monthly forward rates and spot rates of the Australian dollar from May 1991 to April 1997, a transfer function model is specified with { f t }, the first difference of the 90-day forward rate measured as U.S. cents per Australian dollar, as the input series; and {st }, the first difference of the spot rate, as the output series. Both { f t } and {st } are stationary. i) Describe step by step how you would “prewhiten” { f t } and {st } in the transfer function model. Are the prewhitened series necessarily white noise? (5 marks) ii) Suppose that the sample cross-correlations between the prewhitened { f t } and prewhitened {st } are given by, r fs (5) 0.001 ; rfs(-4) = -0.102; rfs(-3) = 0.004; rfs(-2) = 0.000; rfs(-1) = 0.024; rfs(0) = -0.113; r fs (1) 0.012 ; rfs(2) = 0.130; rfs(3) = 0.516; rfs(4) = 0.202; rfs(5) = 0.193; rfs(6) = 0.101; The standard deviation of r fs (i ) and rsf (i ) is approximately (n i) 1 / 2 , where n is the number of observations in the sample. Discuss how you would identify a tentative transfer function model on the basis of the above results. (5 marks) 2 iii) Suppose that the model identified in ii) is estimated and the Ljung-BoxPierce test for the significance of the residuals are: Q * (6) 31.2 Q (12) 39.5 * Q * (18) 42.4 with p-value = 0.0 with p-value = 0.0 with p-value = 0.0 So what do you conclude? Discuss how you would proceed to specify a final transfer function model. (3 marks) iv) In practice, what other diagnostic tests are necessary to examine the validity of the model? (5 marks) v) Discuss how you would test if there exists significant arbitraging opportunities in the Australian foreign exchange market using the estimated transfer function model. (2 marks) 3 Question 2 (15 marks) a) Briefly discuss the motivation behind log transformation of seasonal time series. Figures 2a and b show, respectively, the earnings per share and log earnings per share of a listed company in the NYSE from 1981Q1 to 2001Q4. Discuss if the objectives of log transformations have been achieved. (2 marks) Fig.2a: Earnings per share 18 16 14 earnings 12 10 8 6 4 2 0 0 10 20 30 40 50 60 70 80 t Fig 2b. Log earnings per share 3 2.5 2 log earnings 1.5 1 0.5 0 0 10 20 30 40 50 60 70 80 -0.5 -1 t b) Denote the log earnings by xt . Upon examining the ACF’s and PACF’s of xt and its first regular and seasonal differences, the investigator estimates the models: i) (1 B 4 )(1 B) xt (1 B)at ; 4 ii) (1 B 4 )(1 B) xt (1 B)(1 B 4 )at ; where at ~ i.i.d .(0, 2 ) . Give expressions of these models in terms of the notation ARIMA(p,d,q)(P,D,Q)L. (2 marks) c) Show that for model ii), the autocorrelation coefficient of the differenced series at lag 3 is given by 3 / (1 2 )(1 2 ) , while for model i), 3 0. (3 marks) d) Results on the estimation of the two models are given as follows. Compare the performance of the models using the in-sample diagnostics. (5 marks) Estimation Results of Model i) Maximum Likelihood Estimation Parameter Estimate Standard Error Approx t Value Pr > |t| Lag MA1,1 0.76654 0.07655 10.01 <.0001 1 Variance Estimate Std Error Estimate AIC SBC Number of Residuals 0.008836 0.093999 -147.514 -145.145 79 Autocorrelation Check of Residuals To Lag 6 12 18 24 ChiSquare 10.80 19.14 22.02 24.79 DF Pr > ChiSq --------------------Autocorrelations-------------------- 5 11 17 23 0.0555 0.0587 0.1838 0.3614 0.169 0.148 -0.119 -0.251 -0.031 -0.014 0.254 0.050 -0.030 0.018 -0.135 0.076 -0.075 0.095 -0.060 -0.065 -0.001 -0.080 0.081 -0.008 0.050 -0.028 -0.119 -0.023 Estimation Results of Model ii) Maximum Likelihood Estimation Parameter Estimate Standard Error t Value Approx Pr > |t| Lag MA1,1 MA2,1 0.68089 0.31449 0.09049 0.11227 7.52 2.80 <.0001 0.0051 1 4 5 Variance Estimate Std Error Estimate AIC SBC Number of Residuals 0.008137 0.090203 -152.753 -148.014 79 Autocorrelation Check of Residuals To Lag 6 12 18 24 e) ChiSquare 3.38 10.17 13.70 16.39 DF Pr > ChiSq --------------------Autocorrelations-------------------- 4 10 16 22 0.4960 0.4256 0.6212 0.7961 0.042 0.150 -0.114 0.187 0.070 -0.089 -0.105 0.114 -0.084 0.025 -0.008 0.070 0.007 -0.053 0.050 -0.121 -0.032 -0.000 0.004 -0.132 0.021 0.097 -0.057 0.027 Obtain forecast of earnings per share for period 85 (i.e., 2002Q1) using the preferred model chosen in part d) and the following information: Obs. 80 81 82 83 84 xt 2.3016 2.7850 2.6858 2.7738 2.4519 x̂t (predicted value of xt ) 2.4816 2.7855 2.8240 2.7707 2.6465 xt xˆt -0.18 -0.0005 -0.1382 0.0031 -0.1946 (3 marks) 6 Question 3 (37 marks) a) Your financial advisor has suggested to you a protective put strategy on your investment: buy shares in a market stock fund and purchase put options on those shares with three month maturity and exercise price of $1040. The stock fund is currently at $1200. However, your professor at university has suggested instead that you buy a three-month call option on the index fund with exercise price of $1120 and buy three-month T-bills with face value of $1120. i) ii) iii) In a table and on a graph, illustrate the payoffs to each of these strategies as a function of the stock fund value in three months. (6 marks) Which portfolio would you think must require a greater initial outlay to establish? (2 marks) Suppose the market prices of the securities are as follows: Stock fund T-bill (face value $1120) Call (exercise price $1120) Put (exercise price $1040) b) $1200 $1080 $160 $8 iv) Make a table of profits realized for each portfolio for the following values of the stock price in 3 months: ST = 0, 1040, 1120, 1200 and 1280. Graph the profits to each portfolio as a function of ST on a single graph. (9 marks) Which strategy is riskier? (2 marks) i) Use the Black-Scholes formula ( C So e T P( Z d1 ) Xe rT P ( Z d 2 ) ), where d1 ln( So / X ) (r 2 / 2)T T and d2 d1 T to find the value of a European style call option on the following stock: Stock price ( S o = 100); Annual interest rate (r = 0.10); Time to expiry (T = 3 months); ii) c) Exercise price (X = 95); Dividend yield ( = 0); Standard deviation ( 0.5 ) (4 marks) Suppose that the standard deviation on the stock increases. Will the option be worth more or less with the higher volatility? (2 marks) In order to better understand the dependence of a security on volatility, an investigator estimates a GARCH(1,1) model for the monthly excess returns of the S&P 500 index ( y t ) from January 1926 to December 1991. Results of estimation using SAS are given as follows: GARCH Estimates 7 SSE MSE Log Likelihood SBC Normality Test 2.70454693 0.00341 1269.46195 -2512.2257 95.0061 Observations Uncond Var Total R-Square AIC Pr > ChiSq Variable DF Estimate Standard Error Intercept ARCH0 ARCH1 GARCH1 1 1 1 0.007453 0.0000818 0.1203 0.8545 0.001547 0.0000238 0.0197 0.0189 1 792 ??? . -2530.9239 <.0001 t Value 4.82 3.44 6.12 45.15 Approx Pr > |t| <.0001 0.0006 <.0001 <.0001 i) Write down the estimated GARCH process. What assumptions on the parameters are necessary in order for the process to be valid? (5 marks) ii) Obtain the unconditional variance of the process. iii) What other diagnostic tests would you use to examine the adequacy of the model? (3 marks) 8 (4 marks) Question 4 (23 marks) To examine the international transmission of bond market movements, an investigator considers 960 daily close-of-trade observations from April 1986 to December 1989 on the yields of government bonds with less than five years to maturity for the bond markets of the U.K. and U.S. a) Using the Augmented Dickey Fuller (ADF) test in conjunction with the Dolado’s sequential testing procedure discussed in class, test for the unit root hypothesis for each of the bond yield series at the 10% level of significance. Some of the following information may be useful: Test Statistic U U UK -1.14 -1.01 US -1.21 -1.73 4.01 6.17 3.92 2.22 where is the test statistic for testing H o : 0 vs. H1 : 0 in models without a linear trend; U is the test statistic for testing H o : 0 vs. H1 : 0 in models with a linear trend; is the test statistic for testing H o : 0 vs. H1 : 0, 0 in models without a linear trend; U is the test statistic for testing H o : 0 vs. H1 : 0, 0 in models with a linear trend; and the regression used for testing is: p xt t xt 1 i xt i t (8 marks) i 1 b) Discuss (without having to perform the testing) how the investigator could test if the bond “returns” (as opposed to bond yields) are stationary? (2 marks) c) Suppose that each of the bond yield series is I(1). The investigator then tests if there is a long run relationship between the series using the bivariate Cointegrating Regression Augmented Dickey Fuller (CRADF) test. The results are summarized as follows: CRADF UK/US US/UK p = augmentation level 0 0 -4.27 -3.98 9 Conduct the test at the 10% level of significance. What do you conclude? (2 marks) d) The investigator then constructs and estimates the following Error Correction model: UKt 01 1 zt 1 USt 02 2 zt*1 q m i UKt i j USt j 1t i 1 j 1 m* q* i*UKt i *j USt j 2t i 1 i) ii) iii) e) (5.1) (5.2) j 1 What is the purpose of including the error correction terms in (5.1) and (5.2)? (2 marks) What are the expected signs of the coefficients 1 and 2 ? Carefully explain your answers. (2 marks) Suggest how the investigator could determine the appropriate number of lagged terms in the Error Correction model. (2 marks) One interesting hypothesis to examine is whether movements in one bond market have a tendency to “Granger cause” movements in the other. Discuss how the test of Ganger causality may be conducted. Suppose the test statistic for testing UKt USt is 9.563 and q q* m m* 6 . Conduct the test at the 5% level of significance. (3 marks) f) Given your answers to parts c) and e), discuss if USt is expected to “Granger cause” UK t . (2 marks) 10