def loss_function(x):
"""Loss function for the optimization."""
x_mr = _to_constrained(x[:num_mean_reversion], mr_lb, mr_ub)
x_vol = _to_constrained(x[num_mean_reversion:], vol_lb, vol_ub)
mean_reversion_param = piecewise.PiecewiseConstantFunc(
jump_locations=[], values=x_mr, dtype=dtype)
volatility_param = piecewise.PiecewiseConstantFunc(
jump_locations=volatility.jump_locations(),
values=x_vol,
dtype=dtype)
model_values = swaption.swaption_price(
expiries=expiries,
floating_leg_start_times=float_leg_start_times,
floating_leg_end_times=float_leg_end_times,
fixed_leg_payment_times=fixed_leg_payment_times,
floating_leg_daycount_fractions=floating_leg_daycount_fractions,
fixed_leg_daycount_fractions=fixed_leg_daycount_fractions,
fixed_leg_coupon=fixed_leg_coupon,
reference_rate_fn=reference_rate_fn,
dim=1,
mean_reversion=mean_reversion_param,
volatility=volatility_param,
notional=notional,
is_payer_swaption=is_payer_swaption,
use_analytic_pricing=use_analytic_pricing,
num_samples=num_samples,
random_type=random_type,
seed=seed,
skip=skip,
time_step=time_step,
dtype=dtype)[:, 0]
if volatility_based_calibration:
model_values = implied_vol(
prices=model_values / annuity / notional,
strikes=fixed_leg_coupon[..., 0],
expiries=expiries,
forwards=swap_rate,
is_call_options=is_payer_swaption,
underlying_distribution=UnderlyingDistribution.NORMAL,
dtype=dtype)
model_values = tf.where(tf.math.is_nan(model_values),
tf.zeros_like(model_values),
model_values)
print(x_mr, x_vol, model_values)
value = tf.math.reduce_sum(
(_scale(model_values, target_lb, target_ub) -
scaled_target)**2)
return value
def loss_function(x):
"""Loss function for the optimization."""
x_mr = _to_constrained(x[:num_mean_reversion], mr_lb, mr_ub)
x_vol = _to_constrained(x[num_mean_reversion:], vol_lb, vol_ub)
mean_reversion_param = piecewise.PiecewiseConstantFunc(
jump_locations=[], values=x_mr, dtype=dtype)
volatility_param = piecewise.PiecewiseConstantFunc(
jump_locations=volatility.jump_locations(),
values=x_vol,
dtype=dtype)
model_prices = swaption.swaption_price(
expiries=expiries,
floating_leg_start_times=float_leg_start_times,
floating_leg_end_times=float_leg_end_times,
fixed_leg_payment_times=fixed_leg_payment_times,
floating_leg_daycount_fractions=floating_leg_daycount_fractions,
fixed_leg_daycount_fractions=fixed_leg_daycount_fractions,
fixed_leg_coupon=fixed_leg_coupon,
reference_rate_fn=reference_rate_fn,
dim=1,
mean_reversion=mean_reversion_param,
volatility=volatility_param,
notional=notional,
is_payer_swaption=is_payer_swaption,
use_analytic_pricing=use_analytic_pricing,
num_samples=num_samples,
random_type=random_type,
seed=seed,
skip=skip,
time_step=time_step,
dtype=dtype)[:, 0]
value = tf.math.reduce_sum(
(_scale(model_prices, price_lb, price_ub) - scaled_prices)**2)
return value