siruiji / ml4frm Goto Github PK

Hello,
Thanks for putting the code to the excellent book up in Github. I notice a very small error in the code - I don't know if it is local of whether the error is common to all of the AR model codes but I'm considering the file: Volatility_prediction_GARCH.py:

Do let me know if you disagree with my reasoning regarding the square root!

1. the plots should read:
   plt.plot(np.sqrt(forecast.variance.iloc[-len(split_date):]), label='Volatility Prediction-GARCH')
   where one should take the square root of the arch_model forecat since this model actually forecasts variance, not volatility in order to plot it against the realised volatility.

My revised code:
def load_raw_data(ticker, start_date, end_date):
price = yf.download(ticker, start_date, end_date)['Adj Close']
ret = price.pct_change()[1:] # take the scaling out here
ret.dropna(inplace=True)
return ret

def model_train(ret):
global garch, q, best_param
bic_garch = []
for p in range(1, 5):
for q in range(1, 5):
garch = arch_model(ret, mean='zero', vol='GARCH', p=p, o=0, q=q, rescale=True).fit(disp='off') # add scaling here by adding rescale=True
bic_garch.append(garch.bic)
if garch.bic == np.min(bic_garch):
best_param = p, q
garch = arch_model(ret, mean='zero', vol='GARCH', p=best_param[0], o=0, q=best_param[1], rescale=True).fit(disp='off')
scale = garch.scale # take scale value here
print(garch.summary())

realized_vol = ret.rolling(5).std()
n = 252
split_date = ret.iloc[-n:].index
forecast = garch.forecast(start=split_date[0])
plt.figure(figsize=(10, 6))
plt.plot(realized_vol, label='Realized Volatility')
plt.plot(np.sqrt(forecast.variance.iloc[-len(split_date):] / np.power(scale, 2)), label='Volatility Prediction-GARCH') # divide by scaling squared hereand take the square root
plt.title('Volatility Prediction with GARCH')
plt.legend()
plt.savefig('corrected_plot.png')
plt.show()

Thank you for your work. It would be nice if you could give the relevant data in the code