Bulding Your Own

PyAnomaly is highly configurable and customizable, and you can easily add new firm characteristics or functions. When you make modifications, do not change the original source directly. Rather, add new modules (files) and define subclasses if necessary. This is because the library can be updated in the future, and if you change the original source, you will lose the changes you made when you update the library.

If you want some of your modules to be added to PyAnomaly, please contact us.

Coding Rule

There are a few coding rules we follow and we suggest you follow them when developing your own library based on PyAnomaly.

1. Methods for firm characteristics

   • A method (function) that generates a firm characteristic should have a name of the form c_[characteristic name].

   • Write a single-line docstring that describes the characteristic and its author(s) (year). The docstring content will be displayed when Panel.show_available_functions() is called.

   • Do not make a method that generate multiple firm characteristics. For each firm characteristic, make one associated method. If for some reason, it is efficient to generate multiple firm characteristics in one method, make a private method for it and then make a method for each firm characteristic that calls the private method. Below is an example of generating earnings persistence and predictability. _ni_ar1_ivol() generates both persistence and predictability and c_ni_ar1() and c_ni_ivol() call _ni_ar1_ivol() to generate persistence and predictability, respectively.

   def _ni_ar1_ivol(self):
       """Earnings persistence and predictability"""
   
       fa = self.data
       n = 5  # window size
   
       reg = pd.DataFrame(index=fa.index)
       reg['ni_at'] = fa.ib / fa.at_
       reg['ni_at_lag1'] = self.shift(reg.ni_at)
   
       beta, _, ivol = self.rolling_beta(reg, n, n)
       fa['ni_ar1'] = beta[:, 1]
       fa['ni_ivol'] = ivol
   
   def c_ni_ar1(self):
       """Earnings persistence. Francis et al. (2004)"""
   
       if 'ni_ar1' not in self.data:
           self._ni_ar1_ivol()
   
       return self.data['ni_ar1']
   
   def c_ni_ivol(self):
       """Earnings predictability. Francis et al. (2004)"""
   
       if 'ni_ivol' not in self.data:
           self._ni_ar1_ivol()
   
       return self.data['ni_ivol']
   

2. Temporary variables

   If you need to add temporary variables to Panel.data attribute, use a name starting with underscore, e.g., ‘_temp’. Any column whose name starts with ‘_’ will be considered a temporary variable and deleted during postprocessing.

3. Date format

   For date arguments, PyAnomaly uses a string of ‘yyyy-mm-dd’ format.

4. Naming convention

   PyAnomaly uses UpperCamelCase for class names and lowercase_separated_by_underscores for other names (functions, methods, and attributes).

Performance Tips

Jitted functions (a function with @njit decorator) can be slow at the first call due to compile time. While most jitted functions are cached to minimize overhead, datatools.apply_to_groups_jit() and datatools.apply_to_groups_reduce_jit() cannot be cached as they receive a function as an argument. Therefore, functions that utilize these functions, e.g., datatools.classify(), datatools.trim(), datatools.filter(), and datatools.winsorize(), can be slow at the first call. If these functions are used only once, disabling jit (set_config(disable_jit=True)) may increase the speed.