Amazing returns = superior stock selection strategy + superior in & out strategy

(Simplified) Demonstration of concept: SEL[“QQQ”] + I/O[“In & Out”]
In the following backtest, I combine a simple tech stock selection strategy, via buying the QQQ ETF, with the 3 Nov 2020 version of the “In & Out” strategy, which is one possible in & out-type tactic (see the In & Out thread for more tactics; link above). The backtest is from 1 Jan 2008 to 30 Oct 2020. The total return is 1,723%.

The components seem to integrate nicely: the QQQ alone (no in & out) would have yielded about 515% during the backtest period, while the In & Out strategy without a tech selection (only holding the market, SPY) would have resulted in about 1,100%.

And an additional (simplified) application, combining a yet more specific tech stock selection, the semiconductor industry, with the In & Out algo. Below are the specs for the same backtest period as above.
SEL[“SOXX”] + I/O[“In & Out”]
Total return: 1,686%
SOXX alone (i.e., always in): 559%
Implications
The In & Out combines slightly better with the QQQ selection (see above) than the SOXX selection, improving the returns from 515% to 1,723% versus from 559% to 1,686%.

Just messing around with Flex4 QQQ version.  Changing TLT and IEF to TMF and TYD pumps results up to 3539% return.

Nathan Swenson: Absolutely, and great point! If we really trust our strategy, leveraged products, such as the 3 x leveraged bond ETFs, can substantially boost returns. In the attached backtest, I have taken it one step further and used 3 x leverage for all holdings, the 'in' side and the 'out' side. Of course, one would not put all the money on a highly leveraged strategy like this. The max drawdown is 50%+, so this can be psychologically quite distressing. It may be something for a (small) portion of one's total investment, if one feels comfortable with leveraged products.

3 x leveraged SEL[“QQQ”] + I/O[“In & Out”]
'In' holdings: TQQQ
'Out' holdings: TMF, TYD (as per Nathan's test above)
Total returns: 21,882%

Note: The In & Out is the latest 'lazy trader edition', reducing the total number of trades from above 3,000 (see in the QQQ version above) to 207 ... 207 glorious trades to get us to the 20,000%+ :)
As always, for details regarding the latest versions of I/O strategies, see our discussion here.

Wow!  Well, I wouldn't likely use TQQQ due to the decay, For bonds, my thinking is that holding period is shorter and they should be less volatile resulting in less decay.  Anyway, that was my thinking.  Those results are amazing!

Per comments made in the other "In and Out" thread, perhaps the risk-off asset selection is at least as critical as the risk-on asset selcetion to provide good performance in the future.  

Peter Guenther

If may I ask about your last algo.
You set symbol for long order:

self.HLD_IN = {self.STKS: 1}
but in the code when you want to send an order to market you have:

wt[self.MRKT] = 1
Is it a mistake or is it right? Cause HLD_IN parameter suggests that here you should have:
wt[self.STKS] = 1

By the way:
Great work with algorith.

You can use TIPS instead of bonds (to slightly reduce returns, but it's widely applicable). Then again, the US gov's only option during bad economic times is to lower rates, and it's very unlikely the Fed would decrease rates as a section of the economy was faltering. I've also looked into going long volatility during market downturns, but it hasn't worked well. 

I've attached the global ETF rotation strategy that was modified from the tutorial one on Quantconnect. 

gpw radar Thanks for joining this thread and well spotted! This looks like a bug from combining the In & Out with the QQQ stock selection; will post an update soon.

Using TIPs we achieve similar returns, but the Sharpe is much worse. 

Corrected: 3 x leveraged SEL[“QQQ”] + I/O[“In & Out”]
'In' holdings: TQQQ
'Out' holdings: TMF, TYD (as per Nathan's test above)
Total returns: 15,438%

Not quite the 20,000%+, still some way to go :)
Thanks to gpw radar for spotting a bug in the earlier code which I think resulted in holding the SPY and TQQQ in parallel, i.e. on margin / leveraged. In future editions, I will try and record the leverage  so that similar issues pop up quickly,
At least the max drawdown also decreased from 50%+ to 40%+, still quite steep of course.

@Joshua Tsai: Thanks for sharing these results regarding TIPs, a long volatility strategy, and the ETF rotation outline. Much appreciated and great thinking!

Thanks for sharing the code. I used your last version ( Corrected: 3 x leveraged SEL[“QQQ”] + I/O[“In & Out”] ) in a small backtest from 2019-now. However, based on the orders I can not see any "In" or "Out" orders. The strategy buys TQQQ, TMF, and TYD and sells it in the same time frame. Despite all this, the performance looked good. I still wonder whether this was intentional.

Based on corrected version and only using TQQQ and TMF, no TYD.

Almost qualifies for competition with nearly 80% alpha score.  Pretty tough for anyone to beat 23000% return over this period!

Large drawdown of course, but otherwise good numbers.

Nathan Swenson Which version did you used? Did you checked the orders?

Great series of tests there Nathan Swenson, thanks for sharing! You cracked the 20,000% again, nice :)

Tien Duy Vo: Thanks for joining the discussion! I am not 100% whether you looked at the order data below, this is from the Corrected algo version above running from 1 Jan 2008 to today. I have copied in a part of 2019. What should be happening is that the algo alternates between holding the leveraged bond ETFs (TMF and TYD) versus holding the leveraged tech stock selection (TQQQ). For example, see the first lines with the negative quantities (-70701 and -29249): it's selling the bonds on 18 Jan 2019 which it was holding before. In parallel, it's buying the TQQQ (see positive quantity 55093). Then on 24 Jun 2019, it's selling the TQQQ quantity (-55093) and it's investing the money in the bonds (see positive quantities). So, the algo sometimes holds the TQQQ and sometimes the bonds, based on the in & out indicator (see the variable self.be_in in the code). Not sure whether this answers the question?

(Sidenote: there are some "invalid" entries in there, were it doesn't seem to be able to get a price for TQQQ. However, this doesn't seem to affect the holdings.)

Peter Guenther : Thanks for the clarification. You are right. If I start the algo on 01/01/2008, I get the same result as you have posted. However, if I start the algo on 01/01/2018 then I get these results.

Fill: $48.15361309433875 USD

Fill: $20.750885826 USD

Fill: $42.892753618 USD

Fill: $20.760609765 USD

Fill: $42.60804548 USD

Fill: $55.87096473697939 USD

Ok, I see that this strategy works much better with SPY derivatives rather that NQ.  Just switching over to SPXL in tandem with TMF (no TYD) I get nearly 40,000% return: