Week of 2013 JL 26

IWM fell by 0.2% this week.  My account fell 0.1%.  The loss-floor has risen to -8.5%.

I converted eleven years of hourly charts to a fresh database of IWM historical prices.  This should yield better results, as now the database is consistent (all downloaded on the same day) and much larger (the old database went back only to 2008, which is not enough to see all the market’s moods).

End of week allocations:

Daily % gain size │ Max loss Sym  Buy │  Fri  Mon  Tue  Wed  Thu  Fri │ Beg   End   FNSR JL15 │ +0.1  10  +0.2  10  +0.1  10  +0.2  10  +0.2  10  +0.2  10  │ -1.5 -1.4 UWM JL15 │ +0.2  10  +0.3  10  +0.2  10  +0.1  10  +0.2  10  +0.2  10  │ -0.6 -0.5 URTY JL15 │ +0.3  10  +0.4  10  +0.4  10  +0.1  10  +0.4  10  +0.3  10  │ -1.3 -0.8 AGQ JL23 │ +0.1  10  -0.1  10  -0.1  10  -0.2   9  │ -1.2 -0.8 SRTY JL24 │ -0.1  10  -0.1  10  -0.1  10  │ YHOO JL25 │ +0.2  10  +0.1  10  │ -0.7 -0.7 IWM │ +26.4     +26.7     +26.6     +25.5     +26.7     +26.2     │ me │ -5.4 -5.2 -5.1 -5.8 -5.2 -5.5 │ floor │ -9.3 -9.1 -9.9 -9.5 -9.0 -8.5 │

FNSR: First completed trade using the new trading model!  Sold it for 3% more than I paid for it, but had invested only 10% of my money, so (with trading costs) gained only 0.2% from it.

Double-TRIX trading model

Now supports short-selling models.

	Long	Short
Buy signal:	When all are true: TRIX(q) > its EMA(2) Either TRIX(q) < 0 OR TRIX(s) > its EMA(2) close ≤ SMA(q) Low > stop For a daily trade, the signal is given after the daily close and so the purchase is at tomorrow’s opening price.  For hourly, signals are given on the hour 11am‥3pm and the purchase is immediate.	When all are true: TRIX(q) < its EMA(2) Either TRIX(q) > 0 OR TRIX(s) < its EMA(2) close ≥ SMA(q) High < stop For a daily trade, the signal is given after the daily close and so the purchase is at tomorrow’s opening price.  For hourly, signals are given on the hour 10am‥3pm and the purchase is immediate.
Sell signal:	When all are true: TRIX(q) < its EMA(2) TRIX(s) < its EMA(2) For a daily trade, the signal is given after the daily close and so the sale is at tomorrow’s opening price.  For hourly, signals are given on the hour 11am‥3pm and the sale is immediate.	When all are true: TRIX(q) > its EMA(2) TRIX(s) > its EMA(2) For a daily trade, the signal is given after the daily close and so the sale is at tomorrow’s opening price.  For hourly, signals are given on the hour 10am‥3pm and the sale is immediate.
Retest sale:	When exactly w periods have passed after a purchase, if close < m times the purchase price, then sell (next day for daily, immediately for hourly).	When exactly w periods have passed after a purchase, if close > m times the purchase price, then sell (next day for daily, immediately for hourly).
Stop price:	If close > SMA(s): stop = CHANDELIER(c⁺,a⁺) Otherwise: stop = CHANDELIER(c⁻,a⁻) For daily or hourly trades, the stop is recalculated once per day, after market close, and is used as the stop-price for the following day.  At any time, if low < stop, then we recognize that the shares were sold and the trade is over.	If close < SMA(s): stop = CHANDELIER(c⁺,a⁺) Otherwise: stop = CHANDELIER(c⁻,a⁻) For daily or hourly trades, the stop is recalculated once per day, after market close, and is used as the stop-price for the following day.  At any time, if high > stop, then we recognize that the shares were sold and the trade is over.

I have added some additional tickers to the set that this model can work with.  SRTY will now be traded using this model rather than the MACD-based one.  Also, the URTY model has been retuned since now there is more archival IWM data to train it on:

		(daily)								(hourly)
Model ticker:		ATML	FNSR	YHOO	SCHH	IWC	SLV	IWM		IWM	IWM
Trade ticker		ATML	FNSR	YHOO	SCHH	IWC	AGQ (2×)	UWM (2×)		URTY (3×)	SRTY (3×)
TRIX(q)		3	3	2	4	14	4	12		16	5
TRIX(s)		17	17	28	23	64	23	51		97	6
SMA(q)		9	1	10	8	4	1	3		1	1
SMA(s)		240	270	120	210	190	250	130		180	220
CHANDELIER(c⁺,a⁺)		27, 4.9	25, 6.0	27, 5.0	10, 5.0	19, 4.7	11, 4.6	32, 3.4		314, 12.9	305, 11.1
CHANDELIER(c⁻,a⁻)		21, 4.2	22, 3.8	21, 3.9	20, 2.3	22, 2.7	20, 2.3	23, 2.7		32, 11.1	24, 3.9
RETEST(w,m)		4, 1.01	3, 0.99	3, 0.995	3, 1.01	4, 0.995	1, 0.99	3, 1.00		999, 1.0	999, 1.0

The RETEST function doesn’t seem to be useful for hourly trading, so I’ve set its parameters to (999,1.0) to disable it for those cases.

IWC is currently on a “buy” signal, but I haven’t bought because it looks like the uptrend is nearly over.  I’m waiting for some retrace action before beginning to use this model.

I have done a test to see whether it really makes sense to model one ticker while trading another.  Here is a chart of simulated gains:

Year	AGQ	SLV (2×)	Model SLV, trade AGQ
2010	1.0%	4.4%	4.7%
2011	8.2%	9.5%	9.8%
2012	1.6%	2.7%	2.7%

As you can see, modelling AGQ directly doesn’t work very well because the leverage interferes with the technical indicators.  Trading a double-sized tranche of SLV is okay (except it uses up twice as much capital), but best results are obtained when modelling SLV while trading AGQ.

MACD-based short-swing/news trades

This model is now out of use.  Next week, I will start using the new double-TRIX model to trade SRTY.

PPO-based swing trades

This model should be retuned now that I have a lot more archival data available.  The TNA and TZA trades haven’t been doing very well with their current tunings.

Gold trading

No “Buy” signal yet for this model, but getting close!  Maybe next week.

Stock-trading robot

Some glitches this week, but it seems the stop-update for daily trades is now working.

Ticker Symbol		Buy date			Buy price			Sell date			Sell price			Acct Profit
		Model	Actual		M	A		Model	Actual		M	A		M	A
FNSR		MA 29 09:30	JL 15 09:30		$12.81	$18.48		JL 25 09:30			$18.67			4.6%	0.2%
UWM		JL 05 09:30	JL 15 09:30		$63.31	$67.95		(Not yet)
URTY		JN 25 10:00	JL 15 09:30		$50.02	$63.70		(Not yet)
AGQ		JL 23 09:30			$18.08			(Not yet)
SRTY		JL 24 15:00	JL 24 14:00		$16.06	$16.10		JL 25 10:00	JL 24 14:00		$15.97	$16.08		-0.1%
YHOO		JL 25 09:30			$27.73			(Not yet)

FNSR: Got very little profit because I was very late to a trade-already-in-progress.  Still, got a profit!

URTY: The new model parameters would have bought on JN 25 rather than JL 01.

SRTY: Last trade using the old MACD-based model.  The robot screwed up and set a stop price that was above the purchase price, so Schwab sold it immediately.  I think this robot problem affects all short hourly models, so it needs looking into.  Luckily, the trade turned out to be a loser so there was no excess loss from aborting it.

Week of 2013 JL 19

IWM rose by 2.0% this week, to its highest value ever.  My account rose 0.7%.  The loss-floor has dropped to -9.3%.

End of week allocations:

Daily % gain size │ Max loss Sym  Buy │  Fri  Mon  Tue  Wed  Thu  Fri │ Beg   End   FNSR │ +0.1  10  +0.0  10  +0.0  10  -0.0  10  +0.1  10  │ -1.7 -1.5 UWM │ +0.1  10  -0.0  10  +0.1  10  +0.2  10  +0.2  10  │ -0.9 -0.6 URTY │ +0.1  10  -0.0  10  +0.1  10  +0.3  10  +0.3  10  │ -1.5 -1.3 IWM │ +24.4     +25.5     +24.9     +25.4     +26.2     +26.4     │ me │ -6.1 -5.8 -6.0 -5.8 -5.6 -5.4 │ floor │ -6.1 -10.1 -9.8 -9.7 -9.5 -9.3 │

Double-TRIX trading model

The TLINE indicator doesn’t seem to work well (the longest steepest trendlines should be the most reliable ones, but they aren’t), so I have removed it.  Meanwhile, I still need something to avoid horrible results for ATML in the years 2000 and 2004.  Instead, I have added two new tests: “SMA(q)” and “Retest sale”.

Buy signal:

When all are true: TRIX(q) > its EMA(2) Either TRIX(q) < 0 OR TRIX(s) > its EMA(2) close ≤ SMA(q) Low > stop For a daily trade, the signal is given after the daily close and so the purchase is at tomorrow’s opening price.  For hourly, signals are given on the hour 11am‥3pm and the purchase is immediate.

Sell signal:

When all are true: TRIX(q) < its EMA(2) TRIX(s) < its EMA(2) For a daily trade, the signal is given after the daily close and so the sale is at tomorrow’s opening price.  For hourly, signals are given on the hour 11am‥3pm and the sale is immediate.

Retest sale:

When exactly w periods have passed after a purchase, if close < m times the purchase price, then sell (next day for daily, immediately for hourly).

Stop-price:

If close > SMA(s): stop = CHANDELIER(c⁺,a⁺) Otherwise: stop = CHANDELIER(c⁻,a⁻) For daily or hourly trades, the stop is recalculated once per day, after market close, and is used as the stop-price for the following day.  At any time, if low < stop, then we recognize that the shares were sold and the trade is over.

Here are the revised parameter values for the first four trades:

	ATML (daily)	FNSR (daily)	IWM (daily)	IWM (hourly)
Trade ticker	ATML	FNSR	UWM (2×)	URTY (3×)
TRIX(q)	3	3	12	75
TRIX(s)	17	17	51	429
SMA(q)	9	1	3	1
SMA(s)	240	270	130	580
CHANDELIER(c⁺,a⁺)	27, 4.9	25, 6.0	32, 3.4	360, 12.9
CHANDELIER(c⁻,a⁻)	21, 4.2	22, 3.8	23, 2.7	110, 12.0
RETEST(w,m)	4, 1.01	3, 0.99	3, 1.00	21, 0.99

For the new RETEST function, w is always around 3 — maybe the 4 value for ATML is an artifact?  The m value is always around 1.00 — maybe it should always be exactly 1.00?

The new SMA(q) function is probably wrong.  Because it says not to buy if price is too high, that makes the system susceptible to missing out on run-ups if price rises too soon and then never falls below the SMA for weeks on end.  For FNSR it seems worse than useless.  For IWM, it nearly doubles the expected return at the daily level, but is useless at the hourly level???  This probably needs more work.

Here are some figures for ATML, showing simulated results for the last 22 years.  The “Hold” figure shows the result for using all my money to hold ATML all year, while “Trade” shows the effect of the double-TRIX trading strategy (using only 10% of my money).  Numbers shown with red background would have been disappointing:

Year	Hold	Trade		Year	Hold	Trade
1992	+87%	-1.0%		2003	+160%	+14.4%
1993	+111%	+4.4%		2004	-35%	-3.0%
1994	+94%	+2.7%		2005	-21%	+5.8%
1995	+34%	+2.1%		2006	+96%	+4.5%
1996	+49%	+2.5%		2007	-27%	-0.4%
1997	-45%	-0.8%		2008	-34%	+0.5%
1998	-17%	+4.1%		2009	+50%	+2.6%
1999	+286%	+7.1%		2010	+183%	+0.4%
2000	-21%	-0.1%		2011	-34%	-2.4%
2001	-33%	+3.0%		2012	-2%	+1.9%
2002	-71%	+9.6%		2013*	+12%	-0.4%

There are three kinds of disappointments: ⑴ ATML makes lots of money, but I get less than half of it; ⑵ ATML makes money but I lose that year [1992, 2013]; and ⑶ I lose 3.0% or more of my money in one year [2004].  For 2013, only the data for the first six months is shown.

Stock-trading robot

So far, so good!  This week the robot correctly updated its stop on URTY each day, and correctly decided not to sell anything.  Next week, I will have it automatically update the stops on FNSR and UWM (which were done manually this week) and it should also be ready to sell those tickers automatically when the trading model says to.

Ticker Symbol		Buy date			Buy price			Sell date			Sell price			Acct Profit
		Model	Actual		M	A		Model	Actual		M	A		M	A
FNSR		MA 29 09:30	JL 15 09:30		$12.81	$18.48		(Not yet)
UWM		JL 05 09:30	JL 15 09:30		$63.31	$67.95		(Not yet)
URTY		JL 01 11:00	JL 15 09:30		$55.03	$63.70		(Not yet)

For FNSR, the opening price for JL 15 was $18.15, but I got a much worse price than that, perhaps because my purchase was not a round multiple of 100 shares.  This suggests that FNSR might not be a good choice for trading because I won’t actually get the prices that the simulator predicts.

For UWM, the purchase date should have been JL 05, not JL 03 as shown last week.

Week of 2013 JL 12

IWM rose by 3.6% this week, to its highest value ever.  My account was unchanged.  The loss-floor remains at -6.1%.

No trades for me this week.

End of week allocations:

Daily % gain size │ Max loss Sym  Buy │  Fri  Mon  Tue  Wed  Thu  Fri │ Beg   End   IWM │ +20.8     +21.3     +22.4     +22.7     +24.2     +24.4     │ me │ -6.1 -6.1 -6.1 -6.1 -6.1 -6.1 │ floor │ -6.1 -6.1 -6.1 -6.1 -6.1 -6.1 │

New double-TRIX trading model

There have been no changes to the robot’s trading models for eleven months (since “Week of 2012 AU 17”).  Since that time, there have been about 60 trades but the value of my account has gone approximately nowhere.  On June 24th, I went to a programmers’ meet-up and showed them some of the robot’s code.  They were flabbergasted that I have continued to work on this thing for 2½ years even though it has never made any money.  What kind of crazy crank must I be to do such a thing???  I decided to create a new model for next month’s meetup, perhaps one that would actually work for a change!

Buy signal:

When all are true: TRIX(q) > its EMA(2) Either TRIX(q) < 0 OR TRIX(s) > its EMA(2) Low > stop There is no trendline from TLINE(t,slopelimit,peakdist,slop) or there have been fewer than numsubpeaks under it. For a daily trade, the signal is given after the daily close and so the purchase is at tomorrow’s opening price.  For hourly, signals are given on the hour 11am‥3pm and the purchase is immediate.

Sell signal:

When all are true: TRIX(q) < its EMA(2) TRIX(s) < its EMA(2) For a daily trade, the signal is given after the daily close and so the sale is at tomorrow’s opening price.  For hourly, signals are given on the hour 11am‥3pm and the sale is immediate.

Stop-price:

If close > SMA(m): stop = CHANDELIER(c⁺,a⁺) Otherwise: stop = CHANDELIER(c⁻,a⁻) For daily or hourly trades, the stop is recalculated once per day, after market close, and is used as the stop-price for the following day.  At any time, if low < stop, then we recognize that the shares were sold and the trade is over.

CHANDELIER(c,a) is a standard technical indicator that I haven’t used before.  We call DONCHIAN(c,high) to find the ceiling, then “hang” our stop below that by subtracting a times ATR(c), the Average True Range for the period.

TLINE(t,slopelimit,peakdist,slop) is a new technical indicator I made up that is supposed to be a simplistic “trendline” generator.  We process each period as follows:

• If there is a trendline and high is below it, the trend continues.  If trend÷high ≤ slop+1 then this is a possible subpeak; if more than peakdist periods have passed since the previous subpeak and/or the trendline-defining peak₁ or peak₂, then increment the number of subpeaks seen for this trend.
• If high > peak₁, this is our new high-point.  It replaces peak₁ and there is no trendline.
• Otherwise, we have a new peak₂.  Compute its slope (on a logarhythmic chart) as ⁱ√peak₂÷peak₁ where i is the number of periods that have elapsed since peak₁.
  • If this “ith root” ≥ slopelimit, the trendline is not steep enough to be usable so high becomes our new peak₁ and there is no trendline.
  • Otherwise, we rescan from peak₁ to count up the subpeaks for the new trendline.
• At the end of each period, if peak₁ was more than t periods ago then it is too old, so replace peak₁ with peak₂ and rescan for a new peak₂ and possible subpeaks.

Here are the parameter values for the first four trades:

	ATML (daily)	FNSR (daily)	IWM (daily)	IWM (hourly)
TRIX(q)	3	3	12	75
TRIX(s)	17	17	51	430
SMA(m)	230	260	130	580
CHANDELIER(c⁺,a⁺)	27, 5.0	25, 6.0	32, 3.7	290, 12.9
CHANDELIER(c⁻,a⁻)	23, 3.8	22, 3.8	23, 2.6	110, 12.1
TLINE(t,slopelimit,peakdist,slop)	23, 0.998, 2, 0.08	35, 0.995, 2, 0.08	56, 0.999, 2, 0.07	210, 0.9999, 3, 0.09
numsubpeaks	1	1	1	2
Trade ticker	ATML	FNSR	UWM (2×)	URTY (3×)

Currently, ATML does not have a “buy” signal but the other three do.  Should I buy now, or wait for the next “sell” and then buy on the following “buy” signal?  Usually I would wait, but Tim Knight wrote that he thinks the market is going much higher this summer, so I told my broker to buy FNSR, UWM, and URTY at Monday’s open.

Ticker Symbol		Buy date			Buy price			Sell date			Sell price			Acct Profit
		Model	Actual		M	A		Model	Actual		M	A		M	A
FNSR		MA 29 09:30	JL 15 09:30		$12.81	???		(Not yet)
UWM		JL 03 09:30	JL 15 09:30		$61.24	???		(Not yet)
URTY		JL 01 11:00	JL 15 09:30		$55.03	???		(Not yet)

Week of 2013 JL 05

IWM rose by 3.3% this week, while my account fell 0.4% to its lowest value ever.  The loss-floor is now -6.1%.

US markets were closed on Thursday for Independence Day.

End of week allocations:

Daily % gain size │ Max loss Sym  Buy │  Fri  Mon  Tue  Wed  Thu  Fri │ Beg   End   SRTY  JL03 │ -0.4  20  -0.4  20  │ -0.8 -0.8 IWM │ +17.5     +19.4     +18.8     +19.1         +20.8     │ me │ -5.7 -5.7 -5.7 -6.1 -6.1 │ floor │ -5.7 -5.7 -5.7 -6.1 -6.1 │

Stock-trading robot

Ticker Symbol		Buy date			Buy price			Sell date			Sell price			Acct Profit
		Model	Actual		M	A		Model	Actual		M	A		M	A
SRTY		(Do not buy)	JL 03 12:00			$19.22			JL 03 13:00			$18.89			-0.4%

SRTY: I failed to notice IWM’s dividend payment Tuesday morning, which caused the robot to make an incorrect purchase Wednesday afternoon.  Fixed.

Week of 2013 JN 28

Week of 2013 JN 28

IWM rose by 1.2% this week, while my account fell 0.7% to its lowest value ever.  The loss-floor is now -5.7%.

End of week allocations:

Daily % gain size │ Max loss Sym  Buy │  Fri  Mon  Tue  Wed  Thu  Fri │ Beg   End   SRTY  JN24 │ -0.3  20  -0.8  19  -0.8  19  -0.8  19  -0.8  0  │ -1.6 -1.0 IWM │ +16.3     +15.6     +15.9     +16.4     +18.2     +17.5     │ me │ -5.0 -5.3 -5.7 -5.7 -5.7 -5.7 │ floor │ -5.0 -6.1 -5.7 -5.7 -5.7 -5.7 │

Stock-trading robot

Ticker Symbol		Buy date			Buy price			Sell date			Sell price			Acct Profit
		Model	Actual		M	A		Model	Actual		M	A		M	A
SRTY		JN 24 10:00			$21.88	$21.87		JN 25 15:00			$21.04			-0.7%	-0.8%

SRTY: Yet another loss.

Week of 2013 JN 21

IWM fell by 1.9% this week, while my account rose 0.3%.  The loss-floor is now -5.0%.

End of week allocations:

Daily % gain size │ Max loss Sym  Buy │  Fri  Mon  Tue  Wed  Thu  Fri │ Beg   End   SRTY  JN20 │ +0.3  20  +0.3  20  │ -1.6 -1.6 IWM │ +18.4     +19.2     +20.6     +19.1     +15.9     +16.3     │ me │ -5.3 -5.3 -5.3 -5.3 -5.0 -5.0 │ floor │ -5.3 -5.3 -5.3 -5.3 -6.9 -5.0 │

Stock-trading robot

Ticker Symbol		Buy date			Buy price			Sell date			Sell price			Acct Profit
		Model	Actual		M	A		Model	Actual		M	A		M	A
SRTY		JN 20 11:00			$20.55			JN 21 10:00			$20.94			+0.38%	+0.33%

SRTY: A gain, for a change!  But not much.

Week of 2013 JN 14

IWM fell by 0.6% this week, while my account fell 1.9%.  The loss-floor is now -5.3%.

End of week allocations:

Daily % gain size │ Max loss Sym  Buy │  Fri  Mon  Tue  Wed  Thu  Fri │ Beg   End   TNA  JN10 │ +0.2  40  -1.1  39  -1.6  38  -1.6  38  -1.6  38  │ -1.7 -2.0 SRTY ¹  JN12 │ +0.2  20  -0.0  20  -0.0  20  │ -1.0 -0.8 SRTY ²  JN13 │ -0.3  20  -0.3  20  │ -0.4 -0.4 IWM │ +19.0     +19.7     +18.4     +17.2     +19.3     +18.4     │ me │ -3.4 -3.2 -4.5 -4.9 -5.3 -5.3 │ floor │ -3.4 -6.7 -6.7 -6.0 -5.3 -5.3 │

Stock-trading robot

Ticker Symbol		Buy date			Buy price			Sell date			Sell price			Acct Profit
		Model	Actual		M	A		Model	Actual		M	A		M	A
TNA		JN 10 11:00			$49.51	$49.64		JN 12 13:00	JN 12 12:00		$47.08	$47.60		-2.0%	-1.6%
		JN 10 12:00			$49.71	$49.72
SRTY ¹		JN 12 12:00			$20.21	$20.22		JN 13 11:00			$20.24			+0.06%	-0.01%
SRTY ²		JN 13 14:00			$19.81	$19.80		JN 13 15:00			$19.58	$19.57		-0.24%	-0.25%

TNA: Bought a tranche, then bought another tranche an hour later because the rebuy-rule immediately applied.  Maybe I should require a delay before rebuying?
      The model says I stopped out, but actually this was a robot-sale due to a slight wobble in price-quotes.

SRTY: Two more losses.  The SRTY² trade was especially silly.  This model seems poorly tuned for the market we’re in these days.

Week of 2013 JN 07

IWM rose by 0.5% this week, while my account fell 0.9%.  The loss-floor is now -3.4%.

End of week allocations:

Daily % gain size │ Max loss Sym  Buy │  Fri  Mon  Tue  Wed  Thu  Fri │ Beg   End   URTY  MA23 │ +0.2  19  +1.2  40  +0.0  39  -1.3  38  -1.3  38  -1.3  38  │ -0.7 -1.4 TZA  JN03 │ -0.7  18  -1.0  18  -1.0  18  -1.0  18  -1.0  18  │ -1.0 -1.0 SRTY  JN05 │ -0.0  20  -0.5  19  -0.5  19  │ -1.6 -1.6 IWM │ +18.5     +19.6     +18.4     +16.8     +18.1     +19.0     │ me │ -0.5 -0.1 -1.6 -3.0 -3.4 -3.4 │ floor │ -1.3 -2.9 -3.0 -4.5 -3.4 -3.4 │

Stock-trading robot

Ticker Symbol		Buy date			Buy price			Sell date			Sell price			Acct Profit
		Model	Actual		M	A		Model	Actual		M	A		M	A
URTY		MA 23 12:00			$108.27	$108.22		JN 05 12:00			$104.84			-1.3%
		JN 03 15:00	JN 03 14:00		$108.76	$108.54
TZA		JN 03 12:00			$32.11	$32.10		JN 05 10:00	JN 05 09:36		$30.56	$30.44		-1.0%
SRTY		JN 05 12:00			$20.62	$20.76		JN 06 10:00			$20.31	$20.29		-0.43%	-0.46%

URTY: Rather unfortunate timing for the purchase of the second tranche, which merely increased the size of my loss.

TZA: Stopped out.

SRTY: Yet another loss.