Shorting bubbles at the top

One idea comes to mind to trigger a boolean or something when the threshold is initially met and then to model an entry point sometime afterwards.  For example the price rises and your 4/30 threshold is met so cross = True then at some point in the future the price is trending down and the 4/30 threshold is met again and thats when you enter your trade.  You may not always time the top perfectly but ideally you would have momentum on your side on the way down.

Another idea could be trading with a buffer so you always have something like 20% of your portfolio in bonds or cash to liquidate to cover margin calls. There's a setting you can use so you never have all of your money in equities, I modified it a bit from what is in the Buy and Hold Futures tutorial in the bootcamp:        

I'll also try to do the opposite for big companies: buy when the stock dips, and sell later. I'll also try breaking up the code into more files so it can be easier to view each class separately.

Hi, AK M

Interesting strategy. Additional ideas:

1. You could try to use short interest (delivered by Quandl) to empower short signals: stocks with high levels of short interest are connected with a high probability of experiencing negative abnormal returns subsequently. Example of using this type of data in QC:self.AddData(QuandlFINRA_ShortVolume, 'FINRA/FNSQ_' + symbol, Resolution.Daily)

    

2. Probably big losses are related with very strong broad market during these trades - did you try to add momentum of a broad market as a filter?

Hey all,

I've incorporated some of your suggestions and have seen improvements. I set the FreePortfolioValuePercentage to 0.2 and made the algorithm more cautious by rebalancing more frequently. I also allocate a variable portion of my portfolio to an index fund such as QQQ during strong markets.

These changes mitigate some of the risk from margin calls I was facing earlier. 

Hi, AK M

Great! Could you please share your backtest, I'll try to implement idea with short interest.

Thank you

Here ya go!

Above is a run with TQQQ. Attached is a run with QQQ.

Hi AK M

thanks a lot for sharing, very much appreciated! I have some ideas for additions. Do you have an updated version? Then I can shere these in your most recent version. Otherwise, I can share in the QQQ version above.

Happy Easter,

Frank

Hey Frank,

I haven't made any changes to this algo since the update with the Qs. Interested in seeing what you have!

Hey AK,

thanks a lot for sharing. I'm new to the community, so studying the logic behind your code is highly instructive. I can't help but wonder a few things

1) To what extent has the performance of this strategy been driven by the performance of a simple long QQQ, and whether we can quantify the edge (I'd like to test this myself but admittedly having started recently the best I can think of is to add a benchmark = QQQ * weight at each day)

2) After such a long-lasting run-up in the index, I wonder how would this behave in a downturn & what's the best way to protect the index from staying in QQQ if it were to nosedive. What would be a reasonable trigger to switch the % of portfolio allocated to it onto holding another index - such as, for the sake of simplicity, bonds?

Dear AK M, please find attached the backtest after some tweaking over the week-end. CAGR is now 115% which seems to be beautiful. However, this is not a tradeable strategy for various reasons, one of them beeing the capacity of the strategy which is evaluated at only $ 4.800. As I am posting this in my coffe-break, only short explanation. But - do you have a github or any other way of contacting you?

Dear LorenzoZC, welcome to QC!! I had implemented a "use bull" switch which is set to "False".

Have fun cloning ;)

Frank, this is great work. It seems that decaying trailing stop losses really increase the profitability. This is something I'll start to incorporate into other strategies. Also, you're very right that the strategy is difficult to trade. It is not scalable to large accounts. I've reached out to support and asked them pass my contact information to you.

Lorenzo, good question. The original strategy doesn't allocate any capital to QQQ. It performs okay, outside of the problems outlined in the OP. The quantify the edge, I think measures such as the Information Ratio, Alpha, Sharpe, and Sortino are your best bet. Also, I think your second point is interesting and worth exploring. In my experience, bonds, currencies, and metals are good havens during market downturns.

AK M, thanks for the nice words ;).

• In the code above, there is a trailing stop loss option which is set to 0 as it did not improve the PSR & CAGR and harmed the win rate for the negative events found in the timeframe starting 2015.
• On the profit side, there is a gradual reduction of the take profit ratio (55%) which kicks in at 70% (32 days) of the maximum position age (45 days) and then reduces take profit (down to 28% at 45 days) to catch volatile securities at their top.
• Part of the increase versus the previous version also results from increasing the position size. There is now a dynamic calculation of the maximum order quantity based on the maximum allowed securities dollar volume which then limits the position size in the later years. I assume that this helps with the structure of the positions found, as the number of positions found is also larger in the later years.

LorenzoZC, another hint to your 2nd question regarding a reasonable trigger to switch QQQ: there is some impressive work if you search for "in out" in the forum, e.g. In-Out continued or quite similar Dual Momentum with out days. Of course, the triggers mentioned there are created from past experiences, but include really interesting macro points of views and structural stock market info.

AK M Frank Schikarski I was curious and decided flip the code from "short the bubble" to "buy the dip". Attached is my backtested result without QQQ or SPY.

As u can see the results are pretty ok, assuming that the code is bug free.

If it works as intended, I suppose this strategy is more like an EMA crossover buy/sell?

Hi everyone!

I tried to test the strategy in Live and constantly got invalid orders with “There is insufficient <papername> available for short sale..” (BSQR and MRIN today for example)

That's ok in general but backtests in QC don't provide information about short availability of the symbol here.

I wonder how backtests and real launches differ in case of this strategy because it doesn't seem to be tradable on some big amount of cash

First, not sure if you saw Alexandre Catarino's post about short availability: 

but that should fix that issue.

Aside from that, the only danger I see with this strategy is if a stock is up heavily for good reason i.e. a merger or acquisition. If the 4 day EMA is twice the 30 day EMA you need to find out why. You could run the strategy semi-manually and just trade on signals, or alternatively you could use Tiingo analysis to flag words like “merger” that would tell the algo to stay away.

Just some ideas

I have expanded on Frank Schikarski 's version, thanks for the clean code I had a lot of fun and learned a lot playing with it.

(I am not a python developer, I might no be following the language's best practices and followed Frank's code style as much as possible)

I added the following to the existing code, which I am not sure if is overrfitting:

• Check for holding.UnrealizedProfitPercent from day 0 in CheckForExits instead of day > 1. (Cause you know, short squeeze can bring a stock to the moon and ruin the portfolio)
• a dynamic TRADE_CUT_LOSS for day 0 or following days
• CheckForStaleOrders function -  cancel all unfilled orders after 24 hours.
• CheckForShortSqueeze function - checks if they price has gone up too fast in a small period of time

And the results were pretty good: 

I noticed a lot of my orders were being rejected by my broker because the stock wasn't available for short sale. So I have implemented Alexandre Catarino's suggestion about short availability, and this is the result with the exact same parameters as before.

Not so good, huh? Also I am not sure how good is Areyu's data. At least one stocks that in the backtest were not available were in fact available in live trading and my order was filled.

At this point I am not sure about the viability of this strategy. I am specially concerned about being victim again of a short squeeze and considering using WallStreetBets dataset to avoid that.

Also, any tips on how to make this strategy's backtest run faster?

Hi Will Urban,
For issues with live trading, please contact support@quantconnect.com with the live deployment attached.We were able to bring the backtest time of the above algorithm down to 21 minutes by doing the following changes:
 

1. Instead of using the loops to create Scheduled events, we created them using a single command for each
2. In the coarse selection, we were looping through the selected symbols multiple times therefore, we reduced this

For improving the backtest time further, we recommend some changes in the strategy logic. We are calling CheckForShortSqueeze and CheckForExits every 3 and 5 minutes, and respectively we can try to check both conditions in the same Scheduled event. Refer to the attached backtest.
Best,
Varad Kabade