How to Generate Insights for Top 10 and Bottom 10

Hi Sheikh

You can save your selected longs and shorts as separate variables to use:

def SelectFine(self, algorithm, fine):
  
        sortedByYields = sorted(fine, key=lambda f: f.ValuationRatios.EarningYield, reverse=True)
        self.forShort = [f.Symbol for f in sortedByYields[:10]]
        self.forLong = [f.Symbol for f in sortedByYields[-10:]]
        return self.forShort + self.forLong

Then for updating your alpha, you can emit insights separately:

[insights.append(Insight.Price(symbol, timedelta(28), InsightDirection.Down)) for symbol in self.forShort]
[insights.append(Insight.Price(symbol, timedelta(28), InsightDirection.Up)) for symbol in self.forLong]
return insights

Cheers
Louis

Hi Louis, thank you for getting back, greatly appreciated!

I am getting an error however:

Runtime Error: AttributeError : 'MyAlphaModel' object has no attribute 'forLong'
  at Update
    [insights.append(Insight.Price(symbol in main.py:line 81
 AttributeError : 'MyAlphaModel' object has no attribute 'forLong'

I put my full code below. I am trying to be a good student and use the Algorithm Framework but if you think I should use a different approach please suggest it.

Can you help me to make the necessary modifications to get it to work pls. Also, in OnSecuritiesChanged(), how can I tell if a security in AddedSecurities should be bought or shorted. In other words, how can I tell if the security allocation in SetHoldings() should be +ve or -ve?

Thanks again!

class UglyRedOrangeChinchilla(QCAlgorithm):

    def Initialize(self):
        self.SetStartDate(2020, 12, 1)  
        self.SetEndDate(2021, 3, 1) 
        self.SetCash(100000)  
        self.SetBenchmark("SPY")

        self.UniverseSettings.Resolution = Resolution.Daily
        self.AddUniverseSelection(
            FineFundamentalUniverseSelectionModel(self.SelectCoarse, 
                                                    self.SelectFine))
            
        self.AddAlpha(MyAlphaModel())
        self.SetPortfolioConstruction(EqualWeightingPortfolioConstructionModel())
        self.SetExecution(ImmediateExecutionModel())
    
    def SelectCoarse(self, coarse):
        
        hasFundamentalData = [c for c in coarse if c.HasFundamentalData]
        filteredByPrice = [c.Symbol for c in hasFundamentalData if c.Price > 10]
        
        return filteredByPrice

    def SelectFine(self, fine):
        sortedByPERatio = sorted(fine, 
                            key=lambda f: f.ValuationRatios.PERatio1YearHigh, 
                            reverse=False)
        self.forLong = [f.Symbol for f in sortedByPERatio[:10]]
        self.forShort = [f.Symbol for f in sortedByPERatio[-10:]]
        return self.forShort + self.forLong
    
class MyAlphaModel:
        
        def Update(self, algorithm, slice):
            insights = []
            
            [insights.append(Insight.Price(symbol, timedelta(28), InsightDirection.Up)) for symbol in self.forLong]
            [insights.append(Insight.Price(symbol, timedelta(28), InsightDirection.Down)) for symbol in self.forShort]
            return insights
    
        def OnSecuritiesChanged(self, algorithm, changes):
            self.changes = changes
        
            for security in self.changes.RemovedSecurities:
                if security.Invested:
                    algorithm.Liquidate(security.Symbol)
                    
            alloc = 1.0 / len(algorithm.ActiveSecurities)
            for security in self.changes.AddedSecurities:
                algorithm.SetHoldings(security.Symbol, alloc)
        

Hi Sheikh

The objects between two classes is not inter-callable. Sadly you cannot super().__init__() the algo class as you need to call AlphaModel for manage class. My recommendation is not to use a separate class for alpha at all.

Cheers
Louis

Thanks Louis. Can we keep this open for a bit, I may have a few other questions.

Sure :)

Hi Sheikh Pancham,
To continue using the Framework for the above strategy, we recommend creating a SymbolData class with attributes as PE ratio and using a dictionary in the AlphaModel to store these data keyed by the symbol. We can update this dictionary in our OnSecuritiesChanged event handler by removing the removed securities and adding the added one, then sorting the securities by their PE score. Refer to the attached backtest. Note that we have changed the logic to calculate allocation. We assumed that you wanted to assign equal weights to securities whose insights are emitted by the AlphaModel. This done by the EqualWeightingPortfolioConstructionModel.
Best,
Varad Kabad

Thanks Varad, I shall try your solution as well. To be honest, I am framework agnostic & have only been trying to use the AF because in the bootcamp that seems to be what is being advocated for. I also don't want to be using what could eventually become a legacy approach, so I'll definitely give your solution a try.

Having said that, I gave the OnData() approach (as suggested by Louis) a try & it seems to be running ok. I also found some code ("for kvp in self.Portfolio") in this forum discussion: 

https://www.quantconnect.com/forum/discussion/5647/fundamental-factor-long-short-strategy-as-an-alpha-model

that I used that allows my strategy to not flatten positions in the Portfolio if they re-appear in the longs & shorts on the next iteration. Otherwise, I'm guessing that would incur additional fees. Attached, pls have a look & let me know if you see any errors.

btw, why do you guys have a candlestick chart for the Strategy Equity? Can you advise how I can change that to see a continuous equity curve chart with the area under the curve shaded?

I want to say also that this is a great platform especially for someone like me who is not a Python expert. It is almost like plug and play. I'm glad I found QC. And the support and community forum is great too!

I have a few other questions but I'll open a new post for those.

Many thanks to both you & Louis!

Sheikh