SPY: Significant improvement to next day returns by tracking XLV/XLU

In part I. you have seen how correlation of two sector ETFs can be used to predict SPY results. In part II. I’ve taken this to a next level and looked at absolute sector performance to improve SPY next day results. I checked a number of different ways. I looked at one sector outperforming another. I looked at the strongest sector only etc. At the end I find it best to look at the performance of a specific sector. The problem looking at specific sector strength, it changes over time. During the recent bull run it’s been important that financials were among the leaders. Where as in the past other sectors have been leading the bull market. Weakness in specific sectors such as healthcare or utilities has been most predictive in identifying good times for the stock markets.  Sector ETFs carry important information in terms of market health.  A lot of money is still invested in long only funds. For these funds the only way to “hide” during bear markets is to park money in defensive sectors (=low volatility and high dividend). Typically these companies are part of the utilities or healthcare sector.

Let me underline that with some facts.

I looked at next day returns in a bull markets (ROC200>0) since 1999. The table has 5 columns. The first showing benchmark results: SPY next day return in a bull market. The second column showing SPY returns after a down day. Column three and four representing SPY performance after a down day in on of the sectors XLV/XLU. The last column showing performance when either XLU or XLV is down.

All returns could be captured with XLV as well as XLU, while all KPIs other than MaxDD% could be improved!

But hold on, we aren’t done yet.

So far this has been an extremely simple test. I looked at more pronounced weakness. Instead of using a down day criteria I applied ROC (9) <0 as a filter to XLV (Healthcare).  Again, further improved results.

Different volatility regimes

Then I thought how might this perform under different volatility regimes. For identifying different regimes I simply used ATR10 and ranked it with PERCENTRANK. Value >0.5 means high volatility environment and <0.5 indicates low volatility environment. Depending on the volatility environment is used different ROC settings (high volatility 9 bars, low volatility 15 bars)

I think the results are good. Please be aware, that i don’t think this makes a trading system. You should rather use this to improve your money management or for timing your entries.

I encourage you to further look into sector ETFs. I’m still not done. Any feedback or comment is highly appreciated.

Amibroker AFL Code

#include ;
#include ;
SetCustomBacktestProc("formulas\\include\\cbt_midlevel.afl",True);
SetOption("CommissionAmount",0.00);
SetBacktestMode( backtestRegular);
SetOption("MaxOpenPositions",1);
SetOption("MaxOpenShort", 1);
SetOption("MaxOpenLong", 1);
SetPositionSize( 100, spsPercentOfEquity );
SetTradeDelays( 0, 0, 0, 0 );
Short = Cover = False;
Buy   = Sell  = False;
ApplyStop( stopTypeNBar, 1, 1, 0, False, 0 );
Vola  = percentrank(ATR(10),252);
Buy   = Year() >= 1999
        AND ROC(Close,200)>0
        AND ( (vola<50 AND ROC(Foreign("XLV","Close"),15)<0) OR (vola>50 AND ROC(Foreign("XLV","Close"),9)<0) );