Example: Zipline with Quandl bundle

[jankrepl]

Unfortunately installing Zipline is a nightmare, only supports Python 3.5.

So maybe investigate open-source backtesters

[turmeric-blend]

bt is a good choice.

[jankrepl]

bt is a good choice.

Have you used it? I know there are some others in the Python ecosystem but there does not seem to be a clear-cut winner, or is there?

[turmeric-blend]

I have played around with bt. I think most of the python opensource libraries fall into two categories, individual asset-by-asset backtest (most of them), and portfolio-based backtest. bt so happen to have both (it also accepts weights as model output, unlike many others), but only because it emphasizes modularity/flexibility. At least this is what I can remember from memory.

[jankrepl]

Cool! I might give it a try! It would be nice to create minimalistic example of deepdow +bt:) Anyway, if by any chance you are interested you can also give it a go:)

[turmeric-blend]

I actually tried awhile back but it seems that to utilize bt well, lots of stuff (in terms the workflow towards plotting/visualizing) in the current deepdow library might have to be changed (which you might not want to), from my perspective at least. One example of bt's visualization is that it just plots the backtested portfolio value (Equity Progression), comparing different strategies (eg. s1, s2) and produces a report, like:

Stat                 s1          s2
-------------------  ----------  ----------
Start                2010-01-03  2010-01-03
End                  2017-02-22  2017-02-22
Risk-free rate       0.00%       0.00%

Total Return         81.30%      40.79%
Daily Sharpe         1.19        1.45
Daily Sortino        1.57        2.00
CAGR                 8.69%       4.91%
Max Drawdown         -7.83%      -4.07%
Calmar Ratio         1.11        1.21

MTD                  2.08%       1.56%
3m                   4.08%       2.66%
6m                   3.26%       0.47%
YTD                  3.11%       2.27%
1Y                   12.04%      5.49%
3Y (ann.)            6.82%       3.97%
5Y (ann.)            8.12%       4.02%
10Y (ann.)           8.69%       4.91%
Since Incep. (ann.)  8.69%       4.91%

Daily Sharpe         1.19        1.45
Daily Sortino        1.57        2.00
Daily Mean (ann.)    8.61%       4.85%
Daily Vol (ann.)     7.23%       3.34%
Daily Skew           -0.35       -0.29
Daily Kurt           3.80        2.87
Best Day             2.48%       1.20%
Worst Day            -3.11%      -1.13%

Monthly Sharpe       1.41        1.68
Monthly Sortino      2.61        2.61
Monthly Mean (ann.)  8.61%       5.04%
Monthly Vol (ann.)   6.10%       3.00%
Monthly Skew         0.01        -0.59
Monthly Kurt         0.18        0.03
Best Month           5.69%       1.91%
Worst Month          -3.39%      -2.09%

Yearly Sharpe        1.62        1.61
Yearly Sortino       -           -
Yearly Mean          7.25%       4.08%
Yearly Vol           4.46%       2.53%
Yearly Skew          0.15        -0.45
Yearly Kurt          -0.71       -0.03
Best Year            14.10%      7.02%
Worst Year           1.17%       -0.13%

Avg. Drawdown        -0.79%      -0.40%
Avg. Drawdown Days   13.31       13.28
Avg. Up Month        1.64%       0.83%
Avg. Down Month      -1.27%      -0.69%
Win Year %           100.00%     85.71%
Win 12m %            96.00%      94.67%

So this is different from your current setup whereby you plot the moving metrics (Sharpe, Mean etc). I ended up doing halfway and gave up. I will dig up what I can when I am free and post it here :)

[jankrepl]

Interesting! I guess one major thing that is missing in deepdow is that it is not dynamic - it does not tell you when to change the weights of the portfolio. It is meant to predict an ideal portfolio at a given point in time and then we should buy it and hold it. Of course one can always just recompute a new allocation after each horizon and then adjust your portfolio accordingly.

However, if we start with simple things, backtesting a buy and hold strategy should be trivial in any backtester, right?