Master thesis: Deep Structural estimation: with an application to market microstructure modelling

This package proposes an easy application of the master thesis: "Deep Structural estimation: with an application to market microstructure modelling"

The figure above shows the log-likelihood value of the PIN model (left) and the Deep-Surrogate (right)

Authors

• Guillaume Pavé (guillaumepave@gmail.com)

Supervisors

• Simon Scheidegger (Department of Economics, HEC Lausanne, simon.scheidegger@unil.ch)
• Antoine Didisheim (Swiss Finance Institute, antoine.didisheim@unil.ch)

Deep surrogate (architecture)

Hyparameter	Value
architecture	[400,400,200,100]
activation function	Swish
optimizer	ADAM
loss function	MSE
learning rate	0.5e-3
# of epoch	15

Instruction

1. Cloning project

git clone https://github.com/GuillaumePv/pin_surrogate_model.git

2. Going into project folder

cd pin_surrogate_model

3. Creating your virtual environment (optional)

python3 -m venv venv

4. Entering in your virtual environment (optional)

• Mac OS / linux

source venv/bin/activate venv venv

• Windows

.\venv\Scripts\activate

5. Installing libraries

• Python 3

pip3 install -r requirements.txt

6. Using this library for your project

• Create a folder called "data", download data from this link: https://drive.google.com/drive/folders/1zNTgVVaENjVuB2ORKXpEsXcdSQbWgkDx?usp=sharing
• Put datasets inside "data" folder
• Instantiate a surrogate object with: surrogate = DeepSurrogate()
• Use get_derivative to get the first derivative of the log-likelihood function's for each input:
  • surrogate.get_derivative(X)
• Use get_pin to get the PIN value with the number of buy and sell trades computed thanks to the Lee and ready algorithm
  • *surogate.get_pin(X) -> X should be a pandas Dataframe containing 'Buy' and "sell colmuns. Or a numpy array with the colmuns in the following order: ['buy', 'sell']
• The Input X should be a pandas DataFrame containing the name of the models parameters. Or a numpy with the columns in the order below:
  • PIN | ['alpha', 'delta', 'epsilon_b', 'epsilon_s', 'mu', 'buy', 'sell']

Example

• To see a demonstration of inverse modelling procedure: see estimate_par_lbfgs.py
• To see how to determine the PIN value: see demo.py and demo.ipynb

Parameter range

Surrogate model are defined inside some specific range of parameter. PIN model in this surrogate library have been trained inside the range defined the table below. The surroate can not estimate PIN probability with parameters outside of this range of parameters.

Parameter	Min	Max
a	0	0.99
δ	0	0.99
μ	100	300
ε_buy	100	300
ε_sell	100	300
# of buy trades	55	700
# of sell trades	55	700

Python library

pip install DeepSurrogatepin

Link of the PyPl library: https://pypi.org/project/DeepSurrogatepin/

If you find a bug or would like to request a feature, please report it with the issue tracker. If you'd like to contribute to this project, feel free to fork it on GitHub.