125.

811
Advanced Risk Analytics

Vietnam semester
2023

Lecturer: Professor Hung Do

Student’s name: Bui Phi Thanh Lam
Student’s ID: 2100180
Part I – Market Risk Modelling (70 marks)

GARCH model and Value-at-Risk (Review Week 4 Practice Note)

1. Get access to Yahoo Finance and download five-year historical daily prices of a stock or stock
index.

2. Divide the sample into two sub-samples: in sample for model estimation, and out-of-sample for
forecasting. The length of the out-of-sample is two weeks (10 trading days)

3. Calculate the logarithm returns for the stock. Analyze the stock return characteristics, including,
return distribution, serial correlation, and volatility clustering for the in-sample period.

I have selected Apple Inc. (AAPL) as the company whose stock will be utilized for conducting tasks
related to the GARCH model and Value-at-Risk in the Market Risk modeling section. The historical
daily price data for Apple Inc. covers a five-year time horizon, ranging from 27/08/2018 to
11/08/2023 for the in-sample period, and from 14/08/2023 to 25/08/2023 for the out-of-sample
period. After calculating the returns based on the stock prices, I will present the key characteristics of
the returns observed during the in-sample period.

Return distribution:

The Jarque-Bera statistic, with a value of 1152.581 for AAPL's stock price returns, indicates a
significant deviation from a normal distribution. This departure from normality is characterized by
left-skewness, with a negative skewness statistic of -0.256, signifying that AAPL's stock price
experienced moderate growth over the past 5 years.

Furthermore, the distribution of AAPL's stock price returns exhibits heavier tails compared to a
normal distribution, as indicated by a kurtosis value of 7.68. This suggests that AAPL's stock price
displayed greater volatility and sensitivity to market changes than what would be expected for a
normally distributed asset.

This heightened sensitivity to market fluctuations is also reflected in the company's beta coefficient,
which stands at 1.29. A beta exceeding 1 implies that the stock is anticipated to have higher volatility
relative to the broader market, making it more responsive to market shifts.
Serial correlation:

The summary of data in the correlogram table below shows that there is a serial correlation within
AAPL's squared stock price returns. The considerably high Q-Stat or the p-value being extremely near
to zero across all lag periods implies this. Each lag is a combined test to see if at least one of the
previous or current delays is statistically significant in terms of autocorrelation (AC). In other words, it
looks for at least one pair of returns with autocorrelation stretching from lag 1 to lag n.

Volatility clustering:

To check if AAPL's stock price returns had periods of increased volatility, known as the ARCH effect, I
used two methods. One method involved looking at a chart of the returns, while the other involved
using a statistical tool called the correlogram to examine how squared returns were related.

Both methods showed that the ARCH effect was present in the stock price returns. Essentially, this
means that there were times when the stock's returns became more volatile. The statistical test I
used gave the p-values that confirmed this. Additionally, when I looked at the chart of AAPL's stock
price returns, we could see this increase in volatility, especially during the early months of 2020
when the COVID-19 pandemic started spreading.
4. Based on analyses in step 3, build an appropriate GARCH model for the return volatility using the
in-sample period and explain why the model specification is appropriate. (Note that: you may not
need to capture the serial correlation of the return level using ARMA model, but if you do so, you can
earn the reward mark of 5 marks).

Residual Squared Lag (-1):

Coefficient (0.101105): This coefficient suggests that there is a positive relationship between the
squared residuals from the previous period and the current conditional volatility of AAPL's stock
returns. In simpler terms, when there was a significant deviation from the expected return (a high
squared residual) in the previous period, it tends to contribute to higher volatility in the current
period's stock returns.

p-value (0.0000): The very low p-value (essentially zero) indicates that this relationship is statistically
significant. In the context of AAPL's stock returns, it means that the squared residuals from the
previous period have a meaningful impact on the current volatility. In other words, large deviations
from expected returns tend to persist and affect future volatility.
GARCH Lag (-1):

Coefficient (0.866296): This coefficient indicates a positive relationship between the previous
period's conditional variance (GARCH term) and the current conditional variance of AAPL's stock
returns. If the stock experienced high volatility in the previous period, this tends to contribute to
higher volatility in the current period's returns.

p-value (0.0000): The very low p-value here also suggests that this relationship is highly statistically
significant. In the context of AAPL's stock returns, it means that the volatility experienced in the
recent past (the GARCH term) significantly impacts the current volatility. This is consistent with the
idea that periods of high volatility tend to cluster together.

Second, we can look at the ARCH LM test (say choose lag 5, that is, jointly test for the serial
correlation from time t-5 up to now)

Prob is also high, so we cannot reject the null hypothesis that there is no ARCH effect

Overall, the GARCH(1,1) effectively captured heteroskedasticity.

5. Fix the model specification you found in Step 4, fix the size of the in-sample period as the window
size, using 1-step (day) rolling window method to re-estimate the model, make 1-step ahead
forecasts of the volatilities, and calculate the 99% Value-at-Risk for the out-of-sample period.

After forecasting GARCH using the rolling window approach, the 99% Value-at-risk ("VAR") would be
determined as equal to -2.33*GARCH estimates. The results of computing the VAR are shown in the
table below.

6. (5 marks) Graph the calculated 99% Value-at-Risk and the observed returns in one figure, calculate
the hits rate (hits =1 if a daily observed return is greater than the calculated 99% Value-at-Risk and
hits =0 otherwise; and ℎ𝑖𝑡𝑠 𝑟𝑎𝑡𝑒=Σℎ𝑖𝑡𝑠𝑁𝑡=1𝑁) and comment about the accuracy of the model.
 Hit rate=100%

Because hits are equal to 1 in all 10 daily observations, regardless of whether the daily observed
return is greater than the calculated 99% Value-at-Risk (VaR), it suggests that the model is excessively
conservative. In other words, the model is consistently overestimating the risk by assuming that all
observations exceed the VaR threshold.

In practice, a VaR model with a hits rate of 1 for all observations would be considered overly cautious
because it implies that the actual losses (daily returns) never breach the risk threshold set by the
model. This would be highly unusual and would indicate that the model's VaR estimate is not
reflective of the true market volatility and risk.

Nevertheless, it would be premature to conclude that the forecasting model is inadequate due to the
limited sample size of 10 observations. To improve the reliability and effectiveness of this model, it is
essential to gather more data and conduct stress tests to obtain a more comprehensive
understanding of its performance.

Dynamic Conditional Correlation – Bivariate GARCH model (Review Week 5 Eviews Practice Note)

1. Using one pair of data (including the spot and futures price) from the given data file on stream (file
name: Data for assessment 2). Look at the first tab of the file for the assigned data.

2. (15 marks) Estimate their dynamic conditional correlation using the DCC-MGARCH (1,1) model.

The target commodity is orange juice.

T(1) + T(2) <1 indicates that the conditional correlation in the models is not constant over time,
stability condition is met.

As a result, we can observe that being a consumer-stable good, which is necessary for consumers’
daily lives, most of the DCC would fluctuate below 1, indicating that the market projects the prices of
such goods will either always grow or be maintained at the spot price even if the market is
experiencing a crisis.

3. (10 marks) Based on the estimated dynamic conditional correlation (𝜌𝑡) and estimated standard
deviation of the spot (𝜎𝑆,𝑡) and futures (𝜎𝐹,𝑡), calculate the dynamic optimal hedge ratio as follows:

ℎ𝑡∗=𝜌𝑡*(𝜎𝑆,𝑡/𝜎𝐹,𝑡 )

The value of ht* below 1 indicates that, based on the time-varying correlation (𝜌𝑡) and
volatility (𝜎𝑆,𝑡 and 𝜎𝐹,𝑡) of the spot and futures returns, the model suggests a reduced hedge
position compared to a full hedge (where ht* would be 1). Market participants are not fully hedging
their risk.
4. (10 marks) Identify at least one variable that significantly drives the dynamic optimal
hedge ratio (Hint: to prove the significance, you can use the OLS regression with the dependent
variable is the optimal hedge ratio)
 Positive Relationship: The positive sign of the coefficient indicates a positive relationship
between the dynamic conditional correlation (RHO12) and the dynamic optimal hedge ratio (ht*).

Hedging Intensity: This suggests that when the spot and futures returns are more positively
correlated, hedgers may choose to take larger hedge positions in the futures market to mitigate
their risk in the spot market. In other words, they may increase their futures exposure when they
expect the spot and futures prices to move more closely in tandem.
Overall, a positive and significant coefficient for RHO12 in the regression model indicates
that the correlation between the spot and futures returns plays a meaningful role in determining the
optimal hedge ratio, and market participants respond to changes in this correlation by adjusting
their hedging strategies accordingly.
 Part II – Credit Risk Modelling (30 marks)
1. Employ the Mortgage.sas7bdat and lgd.sas7bdat datasets downloaded from
http://www.creditriskanalytics.net/datasets.html
2. Build the best Probability of Default model you could and assess its performance. (15
marks)
 Here are the results from our Probability of Default ("PD") model. It's clear that all the
factors we looked at are important when it comes to predicting the likelihood of a company
defaulting on its loans.

FICO Score: If a company has a lower FICO score (a measure of its credit rating for
mortgages), it's more likely to default on its loans. This makes sense because a lower credit score
suggests a higher credit risk.

Loan-to-Value Ratio (LTV): When a company's LTV ratio is higher (meaning they've borrowed
a lot compared to the value of their property), they are more likely to default. This is because it's
harder for them to recover their property if things go wrong.

Economic Performance (GDP): When the economy is doing well, the chances of a company
defaulting on its loans go down. A strong economy means better business conditions.
Interest Rates: If the prime interest rate is higher, it's tougher for companies to meet their
debt payments. So, when interest rates rise, there's a greater risk of defaults.
Unemployment rate: The higher the unemployment rate the higher the default rate because
citizen will not be able to meet the payment.

3. Build the best Loss Given Default model you could and assess its performance. (15 marks)

Loan-to-Value Ratio (%) (LTV):

Positive and Significant Coefficient: The coefficient for LTV is positive and statistically
significant, it indicates that there is a positive relationship between the LTV ratio and LGD. In other
words, as the loan-to-value ratio increases, the LGD, as a percentage, also increases.
 Interpretation: This suggests that higher LTV ratios are associated with larger losses in the
event of default. Borrowers with higher LTV ratios may have less equity in their collateral, making it
more challenging for lenders to recover their funds in case of default.

Purpose (Renting Indicator):

Positive and Significant Coefficient: The coefficient for the purpose indicator variable (e.g., 1
for renting purpose and 0 for other purposes) is positive and statistically significant, it suggests that
loans taken for renting purposes have higher LGD compared to loans taken for other purposes.
Interpretation: This implies that the purpose of the loan (renting in this case) is a significant
factor affecting LGD. Borrowers who take loans for renting might have different default behaviors or
collateral quality, leading to higher losses in case of default.

Natural Logarithm of Recovery Rate (lnrr):

Negative and Significant Coefficient: The coefficient for lnrr (the natural logarithm of the
recovery rate) is negative and statistically significant, it indicates that there is a negative relationship
between the recovery rate and LGD. In other words, as the recovery rate increases, the LGD as a
percentage decreases.
Interpretation: This suggests that a higher recovery rate is associated with lower LGD, which
is intuitive. A higher recovery rate means that a larger portion of the loan amount can be recovered
after default, leading to lower losses.