Algorithmic Trading Video Series #10 | Backtesting & Performance Analysis

Louis,

In your example code, when I replaced "SPY" with "QQQ" on line 7, 
I got an AttributeError: 'NoneType' object has no attribute 'Price'.
Why?

need to check data before accessing it

below code can help

    def OnData(self, data):
        if not self.sma.IsReady or self.spy not in data or self.bnd not in data:
            return
        if data.Bars.ContainsKey("QQQ"):
            price = data.Bars[self.spy].Price
        
            if price >= self.sma.Current.Value:
                # Either rebalance or rice has crossed above SMA
                if self.Time >= self.rebalanceTime or not self.uptrend:
                    self.SetHoldings(self.spy, 0.8)
                    self.SetHoldings(self.bnd, 0.2)
                    self.uptrend = True
                    self.rebalanceTime = self.Time + timedelta(30)
            # Either rebalance or price has crossed below SMA
            elif self.Time >= self.rebalanceTime or self.uptrend:
                self.SetHoldings(self.spy, 0.2)
                self.SetHoldings(self.bnd, 0.8)
                self.uptrend = False
                self.rebalanceTime = self.Time + timedelta(30)
            
            self.Plot("Benchmark", "SMA", self.sma.Current.Value)

Spacetime 
Thank you for your response.
Why the original data check 
    if not self.sma.IsReady or self.spy not in data or self.bnd not in data: return
is good for "SPY"
or if I use price = self.Securities[self.spy].Price?

I do not know the reasons why. Maybe someone more knowledgeable can shed some light.

Love the series, keep up the good work!

Hi Vladimir and Spacetime,

It's possible for
 

self.spy not in data

to be false and
 

self.spy not in data.Bars

is still true
Sometimes although the key for a security may be available in the slice data, the bar data for that security may not be available. For example, this may happen when non-price(splits,dividends, etc) data is available for that symbol while bar data is not.
Best,
Varad Kabade

Varad kabade , 

thank you for explaining it

it makes more sense now

Warad Kabade,
which code is more reliable

    if data[self.spy].Price >= self.sma.Current.Value:

or

    if self.Securities[self.spy].Price >= self.sma.Current.Value:

Hi Vladimir,
We recommend using 
 

 if data[self.spy].Price >= self.sma.Current.Value:

instead of
 

if self.Securities[self.spy].Price >= self.sma.Current.Value:

because the latter contains the price at which the asset was traded at last, which may not be ideal if the current data does not contain any TradeBar for the given asset. However, in general, they have the same value, so we can use the latter if we only care about the security price no matter when it was recorded.
Best,
Varad Kabade

Varad kabade,

Thanks for reply

If 

We recommend using 

     if data[self.spy].Price >= self.sma.Current.Value:

instead of

    if self.Securities[self.spy].Price >= self.sma.Current.Value:

then

why I get 

    AttributeError : 'NoneType' object has no attribute 'Price'

if I change on line 7 "SPY" to "QQQ"???

Vladimir , you still have to check data before accessing it

Hi Vladimir,
The above error occurs because even though the current slice Object (data) consists of data for our given symbol, it is non-trade data (Splits, Dividends, Delistings, etc.) hence data.Bars[symbol] is a null-type object. Refer to the attached backtest for the above demo.
Best,
Varad Kabade

Varad kabade

I have never seen such a long data test in any public algorithm.

if not self.sma.IsReady or self.spy not in data or self.bnd not in data:

    return       

if self.spy in data.Splits or self.spy in data.Dividends or self.spy in data.Delistings and self.spy not in data.Bars:

    return

Last two questions:

Why in the original algorithm published by Louis SPY does not issue a warning without this lengthy test? 

Where in the documentation is it explained?

Thanks for doing these videos, really helpful 😊

Hi Vladimir,

Why in the original algorithm published by Louis SPY does not issue a warning without this lengthy test?

This happens because when self.spy in data.Dividends is true, consequently data[self.spy] is None. During the backtest interval in the original algorithm by Louis, there is no event of type Dividend. Therefore, we don't encounter the error. We don't need a long test; I was just describing how the TradeBars and QuoteBars are not the only data event in our Slice object. The following test is enough:

Where in the documentation is it explained?

Please refer to the following doc for information regarding different data events. Refer to the attached backtest.
Best,
Varad Kabade

Hi Louis,

Absolutely amazing video series. Thanks for creating these great tutorials. Are you planning on creating more videos or is the series completed? I would really like to see a tutorial that covers how to write custom modules for the algorithm framework from scratch. It would be great to have a video on how to create a custom class for each framework module (universe selection, alpha creation, portfolio construction, execution, and risk management). Is that something that you are planning on creating? Thanks again for these fantastic tutorials.

Cheers,
Mitch

Hi Mitch

While waiting for Louis’s video series updating, we also have examples of custom modules of our framework. They’re well-explained and could be easily mimicked. Please find them on the Alpha Framework docs’ top boxes on each modules’ page, or in this collection. Also, implementation of their structures could be found here (alpha, portfolio construction, execution, risk management). We encourage members to try and build their own ones. Practice makes perfect!

Best
Louis Szeto