modelx Documentation#
Use Python like a spreadsheet!
Latest Updates#
Note: Visit Discussions for more frequent updates.
6 April 2024: spydermodelx v0.13.6 is released. See spydermodelx v0.13.6 (6 April 2024).
18 February 2024: modelx v0.25.0 is released. See modelx v0.25.0 (18 February 2024).
26 November 2023: spydermodelx v0.13.5 is released. See spydermodelx v0.13.5 (26 November 2023).
23 October 2023: modelxcython is released experimentally. Visit modelxcython development site for more details.
19 August 2023: A new blog post, Enhanced Speed for Exported lifelib Models on https://modelx.io has been published.
19 August 2023: modelx v0.23.0 is released. See modelx v0.23.0 (19 August 2023).
29 July 2023: modelx v0.22.0 is released. See modelx v0.22.0 (29 July 2023). A blog post, New Feature: Export Models as Selfcontained Python Packages on https://modelx.io explains the export feature in more details.
5 May 2023: spydermodelx v0.13.4 is released to support Spyder 5.4.3. See spydermodelx v0.13.4 (5 May 2023).
4 February 2023: spydermodelx v0.13.3 is released to support Spyder 5.4.0  5.4.2. See spydermodelx v0.13.3 (4 February 2023).
4 February 2023: modelx is now listed on awesomequant, a curated list of awesome libraries, packages and resources for Quants.
What is modelx?#
modelx is a numerical computing tool that enables you to use Python like a spreadsheet by quickly defining cached functions. modelx is best suited for implementing mathematical models expressed in a large system of recursive formulas, in such fields as actuarial science, quantitative finance and risk management.
Feature highlights#
modelx enables you to interactively develop, run and debug complex models in smart ways. modelx allows you to:
Define cached functions as Cells objects by writing Python functions
Quickly build objectoriented models, utilizing prototypebased inheritance and composition
Quickly parameterize a set of formulas and get results for different parameters
Trace formula dependency
Import and use any Python modules, such as Numpy, pandas, SciPy, scikitlearn, etc..
See formula traceback upon error and inspect local variables
Save models to text files and versioncontrol with Git
Save data such as pandas DataFrames in Excel or CSV files within models
Autodocument saved models by Python documentation generators, such as Sphinx
Use Spyder with a plugin for modelx (spydermodelx) to interface with modelx through GUI
modelx sites#
Home page 

Blog 

Documentation site 

Development 

Discussion Forum 

modelx on PyPI 
Who is modelx for?#
modelx is designed to be domain agnostic, so it’s useful for anyone in any field. Especially, modelx is suited for modeling in such fields such as:
Quantitative finance
Risk management
Actuarial science
lifelib (https://lifelib.io) is a library of actuarial and financial models that are built on top of modelx.
How modelx works#
Below is an example showing how to build a simple model using modelx. The model performs a Monte Carlo simulation to generate 10,000 stochastic paths of a stock price that follow a geometric Brownian motion and to price an European call option on the stock.
import modelx as mx
import numpy as np
model = mx.new_model() # Create a new Model named "Model1"
space = model.new_space("MonteCarlo") # Create a UserSpace named "MonteCralo"
# Define names in MonteCarlo
space.np = np
space.M = 10000 # Number of scenarios
space.T = 3 # Time to maturity in years
space.N = 36 # Number of time steps
space.S0 = 100 # S(0): Stock price at t=0
space.r = 0.05 # Risk Free Rate
space.sigma = 0.2 # Volatility
space.K = 110 # Option Strike
# Define Cells objects in MonteCarlo from function definitions
@mx.defcells
def std_norm_rand():
gen = np.random.default_rng(1234)
return gen.standard_normal(size=(N, M))
@mx.defcells
def stock(i):
"""Stock price at time t_i"""
dt = T/N; t = dt * i
if i == 0:
return np.full(shape=M, fill_value=S0)
else:
epsilon = std_norm_rand()[i1]
return stock(i1) * np.exp((r  0.5 * sigma**2) * dt + sigma * epsilon * dt**0.5)
@mx.defcells
def call_opt():
"""Call option price by Monte Carlo"""
return np.average(np.maximum(stock(N)  K, 0)) * np.exp(r*T)
Running the model from IPython is as simple as calling a function:
>>> stock(space.N) # Stock price at i=N i.e. t=T
array([ 78.58406132, 59.01504804, 115.148291 , ..., 155.39335662,
74.7907511 , 137.82730703])
>>> call_opt()
16.26919556999345
Changing a parameter is as simple as assigning a value to a name:
>>> space.K = 100 # Cache is cleared by this assignment
>>> call_opt() # New option price for the updated strike
20.96156962064
You can even dynamically create multiple copies of MonteCarlo
with different combinations of r
and sigma
,
by parameterizing MonteCarlo with r
and sigma
:
>>> space.parameters = ("r", "sigma") # Parameterize MonteCarlo with r and sigma
>>> space[0.03, 0.15].call_opt() # Dynamically create a copy of MonteCarlo with r=3% and sigma=15%
14.812014828333284
>>> space[0.06, 0.4].call_opt() # Dynamically create another copy with r=6% and sigma=40%
33.90481014639403
License#
Copyright 20172024, Fumito Hamamura
modelx is free software; you can redistribute it and/or modify it under the terms of GNU Lesser General Public License v3 (LGPLv3).
Contributions, productive comments, requests and feedback from the community are always welcome. Information on modelx development is found at Github fumitoh/modelx