Trading Strategy: Scaling In/Out of RSI(2)

14Dec08

I struggled with whether or not to share this strategy because of my “never share anything I would trade myself” rule (this one is very close), but in the end I decided to err on the side of the reader.

This strategy will expand on the simple RSI(2) strategy by scaling in and out of positions (as alluded to in my most recent post). To make things more interesting, I’ll assume we used 2x leveraged mutual funds from Rydex or ProFunds (which are inherently frictionless, meaning no transaction costs or slippage).

2008121403
[logarithmically-scaled]

The graph above shows the following strategy (red) vs the S&P 500 (blue) from 2000 to present: go 100% long at today’s close if RSI(2) closes below 5, 75% long on a close below 10, 50% below 15, and 25% below 20. Go 100% short if RSI(2) closes above 95, 75% short on a close above 90, 50% above 85, and 25% above 80.

And for the number lovers:

2008121404

 

Things I like about this strategy:

This strategy has dominated the markets since 2000 in terms of absolute and risk-adjusted returns.

It has performed consistently well over a large number of trades (n=1059), raising my confidence that it will continue to perform in the future. And as the graph below shows, this consistent performance includes both the long (green) and short (red) sides of the strategy, despite some long bear and bull runs in the market.

20081214051
[logarithmically-scaled]

But what I like most about this strategy is how well it has managed losses (including the market collapse this year) with average drawdowns on any given day 90% smaller than a buy & hold strategy. Mind you, it has suffered some sizeable drawdowns (the worst being -28.9% in October of this year), but it has pulled out of all of them very, very quickly.

Things I don’t like about this strategy:

In the end I opted to share this strategy because there are three things that I don’t like about it in its current form: (a) the fact that the RSI(2) indicator didn’t work prior to about 1997 (smell an adaptive strategy coming?), (b) that it is susceptible to the same “abnormal market” collapse that bit my YK Strategy so badly in October, and (c) that it doesn’t account for intermediate indicators (which my quick tests show would even further improve the results here).

I’m reaching my self-imposed word limit for this post, so in a follow up post I’ll discuss those three issues and my solutions for each.

[Edit: click for a summary of posts related to RSI(2)] 

Happy Trading,
ms

To stay up to date with what’s happening at the MarketSci Blog, we recommend subscribing to our RSS Feed or Email Feed.



23 Responses to “Trading Strategy: Scaling In/Out of RSI(2)”

  1. its funny i recently tested something similar that incorporated scaling—-although i actually use a percentile rank n-day return (usually 3 days) vs using the RSI, i found that filtering for more volatile days dramatically improved the strategy, and also when i trade this way (which i have for some time) I simply use index options or vix options—whichever gives the best deal at the time. Using options obviously limits you risk dramatically. During the period where you had a 28% drawdown, i bought the index calls (for about 10% of the portfolio) as you would have bought the ETF and subsequently lost, but i hedged my position by buying a small amount (2%)of the in the money vix 50 call for .50 when the vix was trading close to 55, this option went up almost 20x in value–i cashed out early after a 10 bagger–and the result was i won net . These strange discrepancies show up all the time, so keep an eye on it.

    btw good research

    dv

  2. as a more related revision to your strategy i suggest the following:

    when the market is above the 200d ma you revise the scaling for short trades:

    short 12.5% at 80, 25% at 85, 37.5% at 90, 50% at 95, scale increase of 12.5% each day above 95

    this accounts for the tendency for overshoot and limits position sizing on average relative to more profitable long trades

    when the market is below the 200d ma you revise the scaling for long trades the exact same way, obviously this will help solve the drawdown problems

    if you backtest this let me know what you find

  3. 3 MrX

    When do you exit in the scale-in scenario ?

  4. 4 marketsci

    RE to MrX: when the condition isn’t met. For instance, day 1 RSI(2) closes at 8 and the strategy takes a 75% position at the close, day 2 RSI(2) closes at 4 and the strategy takes a 100% position at the close, and on day 3 RSI (2) closes at 30 and the position moves to cash. Completely random numbers, but hope it helps. michael

  5. 5 Ramon

    So Michael, are you implying that the RSI(2) level would need to estimated prior to close to trade this strategy? Subsequently, would you agree that this could lead to issues with replicating backtested performance?

    Also could you please explain why there is no slippage with the funds mentioned (Rydex etc)? Are you guaranteed the closing price on these?

    Thanks and keep up the great work!

  6. 6 MrX

    Let’s say you scale in below 20,15,10 & 5 … Than you go to cash when it gets above 20 again (first condition to open position)?

  7. great stuff!

    although i can’t say i’m that happy that rsi(2) is getting so much recent exposure on the blogosphere :)

    josh

  8. 8 marketsci

    RE to Ramon: correct. Two options here. (a) you could pretty easily reverse engineer the RSI formula to get each day’s trigger points, or (b) some charting programs chart the RSI in realtime based on whatever the price is at that moment. And I agree, the trader would occassionally get it wrong if the market really moved after the fund cutoff time (about 10 mins before the market closes). michael

  9. 9 marketsci

    RE to MrX: you got it. Apologies for not making that clearer. michael

  10. Michael, there is a very simple solution to the drawdown/collapse issue. However, since the solution would require me giving up some of our “trade secrets,” I will not elaborate, although I suspect what I am referencing is probably going to be covered by your “intermediate indicators.”

    For anyone out there interested, I have been trading in my somewhat audited Covestor account a very simliar ETF strategy, based in large part on RSI2. You’ll notice the equity curve of my account is very similar to the ones Michael presents above

  11. Tested RSI(2) strats in Bovespa (Brazil) and got similar results. However it did not perform at all from 2000 to 2004(bear market). It depends very much on the crossover levels that you choose, higher cross levels (15 for example) yield more trades but more losers.

    Also, I found that using 1/4 of your position and opening new positions every 2 days if RSI(2) hasn’t bounced from its lows will increase profitability, but some ppl may find this too risk as we are adding to a losing position.

    one last coment is that I found that you can get better results if you increase the lenght of the RSI when the close price is below a long term MA, lets say MA(100). So RSI(2) if Close>MA(100) and RSI(3) if Close<MA(100)

    There is a chart in my Blog showing the results (but text is in portuguese)

    Sandor

    http://giuocopiano.blogspot.com/

  12. 12 Cameron

    Sorry to ask such a simple question, but what data do you use before the Rydex and ProShares existed? The relationship between 2*S&P and Rydex or ProShares isn’t strong enough for my taste.

  13. 13 marketsci

    RE to Cameron: good question. Are you looking at the ETFs or the funds? The ETFs are going to have a big track error, but the r-squared on the funds is well over 99%. michael

  14. 14 marketsci

    RE to Sandor: great comment. I hadn’t tested that idea of extending the length of the RSI in a downtrend. Will put it through the paces. Thanks, michael

  15. 15 bhh

    I would have preferred this be handled a bit more discreetly.

  16. 16 marketsci

    RE to bhh: if you’re implying that my humble blog could somehow lead to the arbitraging away of this opportunity, then I say “thank you sir”. I understand your concern, but in all seriousness, folks who drive far more traffic than I have discussed the RSI(2) publicly for quite a while (and yet still it persists).

    What I won’t talk about in detail because I do think they are unique are the three solutions I allude to at the end of the post. In a follow up I’ll treat them at a conceptual level, but not with specific rules.

    michael

  17. 17 mm

    You wrote:
    “(a) the fact that the RSI(2) indicator didn’t work prior to about 1997 (smell an adaptive strategy coming?),”

    I have been playing around with RSI2 while trying to implement a new strategy testing system, so I admit that my analysis could be (way) off… but if it’s not, then this is pretty interesting.

    Rules (basically picked at random): Before 1999:
    take 60% long position if RSI > 50 and RSI < 70
    70% between 70 and 80
    80% between 80 and 90
    100% above 90
    Take a similar approach on the short side.

    After 1999 simply flip the long/short signals.

    Using SPX data from 1950 to now, the equity curve looks pretty good. Intuitively, this makes some sense because buying RSI above 50 and selling below 50 would be a trend following approach with flipping the signal (eg post 1999) would be a mean reversion approach.

    I have not been successful in automating the detection of the change from trending to mean reverting. I would be very interested in seeing if you concur that with that simple change RSI2 can be shown to work for the last 50 year and if so, maybe we can discuss how to improve it more by applying some adaptive strategies?

  18. 18 marketsci

    RE to mm: I think conceptually the rules you’ve laid out above are just fine (I haven’t tested them specifically, but looks good) – two quick notes: (a) in the little bit I’ve done with RSI(2) I haven’t found “shallow readings” to be very predictive – in other words, RSI(2) seems predictive when it is very deep into OB/OS territory, but not so much when it’s near the unchanged mark. (b) you’re right that the hard part is the mechanism for adjusting the trading strategy if/when it stops working as an OB/OS indicator. I have my own approach with the YK and Scotty strategies, but I can’t provide too much guidance here. I can say however to be very careful of curve-fitting here – with so few historical shifts in how RSI(2) has worked historically, I would be very wary of building an adaptive mechanism that was overly curve-fitted to those handful of instances. Good luck!

    michael

  19. 19 Anton

    Michael, wonderful research, thank you!

    I have a general question regarding the mechanics of trading with the Profunds?
    What do you normally do if an “index” value is very close to your reference point (e.g., the last day’s close) at 3.55pm or whatever the cutoff time is?

    Do you pass the trade on that day? Take it assuming the relationship with a reading at 3.55pm will be equally strong as the one at the close? What if in the last few minutes the index moves sharply and at 4pm you know that you are in a trade that does not meet your system conditions? Do you hedge immediately or do nothing? Thanks!

  20. 20 marketsci

    RE to anton: this is a very good question and a perpetual pain in the rump – we only trade two programs that require the trader to make a call at the close, YK and Scotty.

    YK requires the trader to just determine if the close is up or down. If unsure at the cutoff time, we take a sharply reduced average of the up and down positions.

    Scotty requires the trader to determine the level the market closes at (within multiple sets of ranges). If it looks like the market will close near where those ranges begin or end, we’ll take an average of the two ranges.

    Not uber scientific I know, but it happens infrequently enough that it works.

    P.S. all of the audited track records you’ll find on the Strategies page include any “bad” trades on our part (i.e. when we chose the wrong closing direction or closing level).

    All the best,
    Michael

  21. 21 White Cube

    I have run this strategy to see how RIS2 with scaling has performed since 2009.
    On the long side the performance has not been as good as before (with some sharp drawdowns).
    On the short side the equity curve went down for 2 years and seems picking up again since 2011.
    Has the magic RSI gone ? Or is it still working if adaptive rules (percentranking) are used?
    Raphael


  1. 1 Monday links: finance is a confidence game « Abnormal Returns
  2. 2 Using RSI 2 to Trade Leveraged ETFs | Brute Force Attack

Leave a Reply

Please log in using one of these methods to post your comment:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s