grac2h5oh/cmf_ffs

My work on a project "Forecasting financial statements of companies using econometrics and machine learning methods" in the Center of Mathematical Finances.

CMF_FFS

Contents:

• Backstory
  • Problems
  • Solution
• Goals
• What will be done
• Models results
• Used methods and models description
• Soon
• Sources

Backstory

Problems

Modeling of the company's operations is mainly done by people and they can make mistakes.

However experienced economists and analytics could do something different instead of this. Creating a new model takes weeks.

Solution

Let's use machine learning to reduce the error and eliminate the human factor.

Goals

Soon

What will be done

• Basic EDA
• Metrics selection
• Fitting baseline model
• Advanced models
• Cross-validation
• feature selection/engineering
• Interpretation of models results(if possible)

Models results

Models and methods	MAPE	WAPE	MSE
Simple linear regressions for each company	0.74	0.74	1835884
Simple linear regression for each companies	47492190	5.894e+06	4.6e+20
Simple linear regressions for each company(cross_val)	0.923	0.811	1517031
Relaxed Lasso with top 20 features for each company	0.355	0.35	5390113
Relaxed Lasso with top 10 features for each company	0.197	0.19	68719
Relaxed Lasso with top 10 features for each company (cross_val)	0.503	0.416	1278153
Catboost(cross_val)	0.369	0.311	141587

Used methods and models description

soon

Sources

i will use this datasets