Inappropriate Shape of stock.spot Due to Floating-Point Precision

[kano5266]

Hi, I was exploring the no transaction band network and encountered an issue while working with BrownianStock. I set dt = 1/365 and time_horizon = 30/365, and after running the simulation, the generated tensor had the expected shape of (n_path, 31).

However, when I changed the time_horizon to 29/365, the tensor should have had a shape of (n_path, 30), but it still generated a shape of (n_path, 31).

This seems to be caused by the precision of floating-point numbers and the use of math.ceil() in line 140:

pfhedge/pfhedge/instruments/primary/brownian.py

Line 140 in 1fc08c7

n_steps=ceil(time_horizon / self.dt + 1),

Example:

from math import ceil

dt = 1/365
time_horizon = 28/365

print(time_horizon/dt + 1)
print(ceil(time_horizon/dt + 1))

29.0
29

from math import ceil

dt = 1/365
time_horizon = 29/365

print(time_horizon/dt + 1)
print(ceil(time_horizon/dt + 1))

30.000000000000004
31 <--WRONG!

from math import ceil

dt = 1/365
time_horizon = 30/365

print(time_horizon/dt + 1)
print(ceil(time_horizon/dt + 1))

30.999999999999996
31

This issue arises due to the floating-point representation, where time_horizon/dt slightly exceeds 30, causing ceil() to return 31 instead of 30.

Proposed Fix

To address this, we can use the round() function to ensure more precise floating-point calculations. For instance:

from math import ceil

dt = 1/365
time_horizon = 29/365

print(time_horizon/dt + 1)
print(ceil(round(time_horizon/dt, 8) + 1)) # Rounds to 8 decimal places

30.000000000000004
30 <-- Correct

This minor adjustment should fix the tensor shape mismatch. Thank you for considering this, and please let me know if any further adjustments are needed!