Strategy Library Addition: G-Score Factor Investing

What interests me about the original Python version above is that it makes use of the Framework components yet contains a lot of customization for the selection of coarse and fine Universe.   I made a C# version which runs up to twice as fast because it doesn't have to marshal objects from python to C# and back. Because it runs faster it is easier to test further customizations such as using StandardDeviationExecution model and Risk management, as I do here.

Thanks for the example above.  Here, below, is a C# version with slight change for Execution model and added Risk management. The key for the ObjectStore is changed because data saved there it is incompatible with the Python version's ObjectStore data.

Thanks again for the python version.  A few notes about the C# version in the backtest above. 

1. Uncomment a line in SaveData to make it save data to ObjectStore. The backtest will then produce different results each time you run it as data accumulates in the ObjectStore (just like the Python version). Hence, after several backtest runs with it saving each time, it might be a good idea to clear the ObjectStore and run the backtest one last time before running live.
2. It appears the C# version runs about the same speed as the Python, though the C# is having to do a bit more work for StandardDeviationExecution and TrailingStop RiskManagement.

Now let's use the Python version for some questions:  

1. I haven't used it live, and I noticed that when you run it repeatedly in backtesting without clearing the ObjectStore inbetween runs there is always a 3 year gap at the beginning of each backtest.  Will it still have the 3 year gap when I run it live? (maybe the variable quarters being set to zero is the culprit)
2. Because the trading starts 3 years into the backtest, isn't the effective CAGR really closer to 20% than 12%?

Hi Kevin,

1. In the FineFilter funtion, we check that (self.quarters<12 and not self.LiveMode) so that the algorithm can collect 12 quarters of data for the backtests, and this causes the 3 year gap. Furthermore, this same boolean will also mean that there will not be a 3 year gap when running the algorithm live. However, please note that the algorithm only trades once per year on the 4th month, so it may take many months before we see our first trade.
2. Yes, this is a fair observation. Lean doesn't remove years of no trading for its calculations.

Best,
Shile Wen

Sorry to bring up an old thread, but can this be done with other sectors?

So instead of:  tech_ROA_key = TECH_ROA

fin_ROA_key = FIN_ROA

Where would I find the above documentation? I understand I would have to additionally go about changing the coarse filter to look at AssetClassification.MorningstarSectorCode:  FinancialServices as opposed to TechonologyServices

Hi Axist,
You may just comment on the ObjectStore parts (avoid survivorship bias) and change the AssetClassification.MorningstarSectorCode condition to another sector.
Best,
Varad Kabade