safewithdrawal_tensorflow

Safe Withdrawal with Certainty Equivalent Cash Flow and TensorFlow

run for gamma = 8 with

./run_safewithdrawal_08.py

• sets up variables to be optimized
• saves in a pickle file
• repeatedly calls safewithdrawal.py
• reduces learning rate each time

./safewithdrawal.py learning_rate steps gamma fileprefix

• loads variables from fileprefix.pickle
• runs optimization using gamma and learning_rate until no improvement for specified steps
• saves variables and csv files summarizing outcome

Safe Withdrawal with Certainty Equivalent Spending and Tensorflow Aug 2016.ipynb

• Jupyter notebook which allows you to run step by step, includes comments and graphs
• However running for a few hours in Jupyter not recommended, browser or Jupyter tends to hang

Please contact with any comments, pull requests

• Run efficiently on GPU
• Use better optimizer like AdamOptimizer instead of GradientDescentOptimizer
  • adaptive learning rate (vs repeated calls with lower learning rate)
  • momentum so less likely to get stuck.