notebook.community

Edit and run 





In [1]:



    


%matplotlib inline
import numpy as np
import pandas as pd
from weekly_buysell_rnn import WeeklyRebalanceRNN



    


















    






Using TensorFlow backend.

















In [2]:



    


DATA = pd.read_hdf("nasdaq_split-adjusted.hdf5")



    











In [3]:



    


DATA



    


















    
Out[3]:








<class 'pandas.core.panel.Panel'>
Dimensions: 80 (items) x 1762 (major_axis) x 5 (minor_axis)
Items axis: AAL to XRAY
Major_axis axis: 2010-01-04 00:00:00 to 2016-12-30 00:00:00
Minor_axis axis: open to volume

















In [4]:



    


stocks = ["AAPL","AMZN","EA","EBAY", "INTC", "GOOGL","NFLX", "NVDA", "REGN" ,"XRAY"]
stocks.sort()
dta = DATA.loc[stocks,:,:]



    











In [5]:



    


strategy = WeeklyRebalanceRNN(dta) #64-32-16
strategy.backtest()



    


















    






Backtest has begun...
====================
1/149 train, date 2011-01-27
Initial training finished, accuracy: 45.60%
====================
3/149 retrain, date 2011-02-25
Retrain finished, accuracy: 49.60%
====================
6/149 retrain, date 2011-04-08
Retrain finished, accuracy: 53.60%
====================
9/149 retrain, date 2011-05-23
Retrain finished, accuracy: 54.40%
====================
12/149 retrain, date 2011-07-06
Retrain finished, accuracy: 52.80%
====================
15/149 retrain, date 2011-08-17
Retrain finished, accuracy: 53.60%
====================
18/149 retrain, date 2011-09-29
Retrain finished, accuracy: 54.80%
====================
21/149 retrain, date 2011-11-10
Retrain finished, accuracy: 48.40%
====================
24/149 retrain, date 2011-12-23
Retrain finished, accuracy: 57.20%
====================
27/149 retrain, date 2012-02-08
Retrain finished, accuracy: 56.80%
====================
30/149 retrain, date 2012-03-22
Retrain finished, accuracy: 55.60%
====================
33/149 retrain, date 2012-05-04
Retrain finished, accuracy: 46.40%
====================
36/149 retrain, date 2012-06-18
Retrain finished, accuracy: 46.80%
====================
39/149 retrain, date 2012-07-31
Retrain finished, accuracy: 58.00%
====================
42/149 retrain, date 2012-09-12
Retrain finished, accuracy: 57.20%
====================
45/149 retrain, date 2012-10-24
Retrain finished, accuracy: 49.60%
====================
48/149 retrain, date 2012-12-10
Retrain finished, accuracy: 44.00%
====================
51/149 retrain, date 2013-01-24
Retrain finished, accuracy: 55.20%
====================
54/149 retrain, date 2013-03-08
Retrain finished, accuracy: 54.00%
====================
57/149 retrain, date 2013-04-22
Retrain finished, accuracy: 56.00%
====================
60/149 retrain, date 2013-06-04
Retrain finished, accuracy: 52.40%
====================
63/149 retrain, date 2013-07-17
Retrain finished, accuracy: 56.40%
====================
66/149 retrain, date 2013-08-28
Retrain finished, accuracy: 55.20%
====================
69/149 retrain, date 2013-10-10
Retrain finished, accuracy: 50.80%
====================
72/149 retrain, date 2013-11-21
Retrain finished, accuracy: 50.40%
====================
75/149 retrain, date 2014-01-07
Retrain finished, accuracy: 35.60%
====================
78/149 retrain, date 2014-02-20
Retrain finished, accuracy: 45.60%
====================
81/149 retrain, date 2014-04-03
Retrain finished, accuracy: 39.60%
====================
84/149 retrain, date 2014-05-16
Retrain finished, accuracy: 45.20%
====================
87/149 retrain, date 2014-06-30
Retrain finished, accuracy: 48.80%
====================
90/149 retrain, date 2014-08-12
Retrain finished, accuracy: 39.60%
====================
93/149 retrain, date 2014-09-24
Retrain finished, accuracy: 44.80%
====================
96/149 retrain, date 2014-11-05
Retrain finished, accuracy: 43.20%
====================
99/149 retrain, date 2014-12-18
Retrain finished, accuracy: 46.00%
====================
102/149 retrain, date 2015-02-03
Retrain finished, accuracy: 42.80%
====================
105/149 retrain, date 2015-03-18
Retrain finished, accuracy: 38.00%
====================
108/149 retrain, date 2015-04-30
Retrain finished, accuracy: 37.20%
====================
111/149 retrain, date 2015-06-12
Retrain finished, accuracy: 29.60%
====================
114/149 retrain, date 2015-07-27
Retrain finished, accuracy: 26.40%
====================
117/149 retrain, date 2015-09-08
Retrain finished, accuracy: 26.80%
====================
120/149 retrain, date 2015-10-20
Retrain finished, accuracy: 36.80%
====================
123/149 retrain, date 2015-12-02
Retrain finished, accuracy: 27.60%
====================
126/149 retrain, date 2016-01-15
Retrain finished, accuracy: 34.40%
====================
129/149 retrain, date 2016-03-01
Retrain finished, accuracy: 33.60%
====================
132/149 retrain, date 2016-04-13
Retrain finished, accuracy: 33.20%
====================
135/149 retrain, date 2016-05-25
Retrain finished, accuracy: 38.40%
====================
138/149 retrain, date 2016-07-08
Retrain finished, accuracy: 42.80%
====================
141/149 retrain, date 2016-08-19
Retrain finished, accuracy: 44.00%
====================
144/149 retrain, date 2016-10-03
Retrain finished, accuracy: 39.20%
====================
147/149 retrain, date 2016-11-14
Retrain finished, accuracy: 36.80%
====================
Backtest finished after 13575.0 s

We started with accuracy above 50% which is still not enough but result from backtest are quite good. After 2014 we stayed around 40% - strategy can be improved by fitting model with another 250 [500] epochs whenever accuracy fall under 45% [40%]. Backtest lasted 3h 46min so I will left it here without other experiments (trained on CPU).





In [6]:



    


strategy.plot()



    


















    






[*********************100%***********************]  1 of 1 downloaded









    
























In [7]:



    


strategy.weights[-5:]



    


















    
Out[7]:











  
    

      
      AAPL
      AMZN
      EA
      EBAY
      GOOGL
      INTC
      NFLX
      NVDA
      REGN
      XRAY
    

  
  
    

      2016-10-17
      0.357072
      0.0
      0.0
      0.000000
      0.000000
      0.0
      0.268531
      0.374397
      0.000000
      0.000000
    

    

      2016-10-31
      0.090927
      0.0
      0.0
      0.000000
      0.000000
      0.0
      0.832801
      0.000000
      0.076272
      0.000000
    

    

      2016-11-14
      0.000000
      0.0
      0.0
      0.347128
      0.000000
      0.0
      0.000000
      0.347101
      0.000000
      0.305771
    

    

      2016-11-29
      0.000000
      0.0
      0.0
      0.002867
      0.000000
      0.0
      0.091082
      0.906051
      0.000000
      0.000000
    

    

      2016-12-13
      0.000000
      0.0
      0.0
      0.417464
      0.092407
      0.0
      0.000000
      0.490129
      0.000000
      0.000000
    

  




















In [8]:



    


strategy.calculate_annual_returns()



    


















    
Out[8]:











  
    

      
      returns
    

    

      date
      
    

  
  
    

      2012-01-25
      0.556045
    

    

      2013-01-24
      0.786666
    

    

      2014-01-22
      0.421765
    

    

      2015-01-20
      -0.157534
    

    

      2016-01-15
      0.212311
    

    

      2016-12-30
      0.420924

Content source: cicavec/AlgoTradingStrategies

Similar notebooks:

notebook.community | gallery | about