Monday, 15 September 2014

Week of 2014 SE 12

IWM fell by 0.9% this week, while my account fell 0.4%.  The loss-floor has risen to -8.3%.

Next week: The FOMC has their meeting, which often causes market uncertainty (will they print A LOT or A WHOLE LOT?).  The ECB will start their TLTRO, which involves printing up money and handing it out to European banks, who will probably use it to buy US stocks and puff up that market.  Oh, and there’s the independence referendum in Scotland; only one recent poll has said that the pro-independence side might win, but that was enough to cause a nosedive in the value of the UK £.

End of week allocations:

16% net equities, 47% cash USD
% gain
size
Max loss
Sym Buy Fri Mon Tue Wed Thu Fri Beg End
YHOO JL22 +2.4 17 +3.3 18 +2.9 18 +3.0 18 +3.1 18 +3.7 19 +0.9 +2.0
URTY AU25 +0.0 14 +0.1 14 -0.4 14 -0.1 14 +0.1 14 -0.4 13 -0.8 -1.5
UWM SE04 +0.0 14 +0.1 1 -0.3 14 -0.2 14 -0.2 14 -0.2 14 -0.7 -0.7
TZA SE09 +0.0 11 -0.2 10 -0.4 10 -0.4 10 -0.6 -0.4
SRTY ¹ SE10 -0.2 10 -0.2 10 -0.2 10 -0.5 -0.5
TNA SE11 -0.0 11 -0.3 10 -0.5 -0.5
SRTY ² SE12 -0.1 11 -0.4 -0.4
IWM   +1.1   +1.2   +0.0   +0.6   +1.3   +0.2
me -5.8 -4.8 -6.0 -6.0 -5.9 -6.2
floor -8.9 -8.6 -8.9 -9.1 -9.3 -8.3

YHOO: Doing great!  But Alibaba’s IPO next Friday might be the news that sets off a “buy the rumour, sell the news” situation with Yahoo.  I will let my robot decide when to sell.  As of now, I have a “guaranteed” +2.0% gain for my account!

UWM, TZA, SRTY, TNA: Whiplash!  Somehow I should find a way to make my system lose less money during these “indecisive” periods.

Stock-trading robot

Ticker
Symbol
Buy dateBuy priceSell dateSell priceAcct Profit
ModelActualMAModelActual M A M A
YHOO JL 22 09:30 $33.48 $33.56 (Not yet)
URTY AU 25 12:00 $87.01 $87.11 SE 12 15:00 $85.01 $84.07 -0.38% -0.41%
UWM ¹ SE 04 SE 04 15:11 $85.29 $85.30 SE 10 09:30 $83.96 $84.02 -0.24% -0.23%
TZA SE 09 15:00 $??.?? $14.67 SE 11 15:00 $??.?? $14.21 -0.34% -0.35%
SRTY ¹ SE 10 10:00 $??.?? $37.99 SE 10 13:00 $??.?? $37.45 -0.14% -0.18%
UWM ² SE 11 09:30 (Skipped) $??.?? (Not yet)
TNA SE 11 15:00 $??.?? $76.66 (Not yet)
SRTY ² (Do not buy) SE 12 14:00 $37.57 (Not yet)

UWM: Sold Wednesday, due to failed retest (price wasn’t up by at least 0.5% after three days).  But conditions were still favourable so the robot generated a “Buy” signal for Thursday, which I decided to skip.

SRTY ²: Due to quote-glitch from Schwab, the robot bought this even though the corrected data at end-of-day says that STDEV(5) wasn’t actually above its EMA(12).  I have added an override rule to the robot so it realizes that I currently own this and need to select an exit-time.

Channel-based trading model

I tried creating a new trading model, but I don’t think it works well enough to use.
  • Buy when price > high of last donchian-H hours, on the hour but not at 10am or 4pm.
  • Sell when price < low of last donchian-L hours.  Update this stop every hour.

Results were okay, but then I tried using my PEAKFREQ(140,8) indicator to choose the Donchian-channel widths.

PEAKFREQdonchian-Hdonchian-LResults 2003‥2011Results 2012‥2013
5155546.1%6.7%
666110
713118
820121
91062
101831
111449
121346
13929
14434
15450
16556
single1412126.3%12.4%

This is not right.  Note that the donchian-H and donchian-L parameters do not progress smoothly as PEAKFREQ is varied.  This implies that the fundamental relationship (if there is one) is getting swamped by random glitches where certain “lucky” values for the parameters happen to have done much better in the past than they will in the future.  I had hoped that using nine years of data for the back-test would reduce this effect, but apparently it’s not enough.

Also note that tuning the parameters by PEAKFREQ nearly doubles the gains for the back-test period, but cuts gains in half for the forward-test period.  Based on these results, I am not justified in making any prediction as to how this algorithm would perform if placed into production.  Oh well.

No comments:

Post a comment