IWM rose by 1.1% this week, while my account rose 0.4%. The loss-floor has fallen to -8.4%.
World news: Politicians can’t seem to decide whether Russia’s actions in Ukraine should be called an ”incursion” or an “invasion”. Either way, it’s open warfare and the EU is likely to get pulled into it soon.
|End of week allocations:
59% equities, 58% cash USD
YHOO: Continuing to do well!
URTY: Looks like a poorly-timed entry, so it didn’t help my profit any this week.
TRIX-based trading model
Things have not been going well. I make less than the market when it is going up and lose more than market on downswings. Why does this keep happening? It’s easy enough to just “blame the Fed” for why my algorithms perform so well on paper and just keep losing when I bet actual money on them, but that response doesn’t solve anything. Yes, it’s true that market dips in recent years have been unusually shallow compared to historical norms, because(?) the central bank keeps printing up a shitstorm of money any time it thinks the market might actually go down for a change. But why can’t my system adjust for that?During the week of JL 18, ⅓ of my loss came from AGQ, which tracks the price of silver. A tighter stop would have prevented this, but back-tests indicate that a tighter stop would reduce profit overall. Silver often has big drops followed by recoveries, but there was no recovery this particular time. So maybe I just have to live with this.
Another ¼ of the loss came from two loser URTY trades, which track the broad small-caps market. The TRIX(q) indicator was wobbling around zero and giving spurious ”buy” signals. But in order to get quick entries when the market is about to go up, I also have to take quick entries where the market doesn’t go up after all. This would be okay if there weren't a perfect storm of other losses the same week.
During the period AL 17‥MA 16, I had nothing but losses over and over. The market was oscillating back and forth and I kept betting it would go one way but then it went the other. Once again, the biggest culprit seems to have been URTY. In particular, the TRIX(s) indicator was wobbling around zero and giving spurious “hold” signals for trades that should have been sold. There *is* something I can do about this: I can bump up the second parameter to TRIX to require a more significant positive value before recognizing a ”hold” signal. This reduces spurious signals but introduces a delay before a valid signal can be recognized. How can I tell if the reduction in risk is worth the reduction in profit?Previously my calculation for ”daily risk” was
( max_loss / purchase_price ) - 1
But this says that AGQ was low-risk during JL 18 because the max-loss was above original purchase price; actually it was high-risk because I am charting week-to-week changes and the current price that week was far above my max-loss level. Another problem: this formula gives negative numbers when things are going well, which makes it hard to figure out how to optimize my parameters, so for goal-seeking I used
( total_gain + average_daily_risk ) / total_gain
but I have no idea whether this math actually makes any sense — and I suspect it doesn’t.
I changed my daily-risk formula to (max_loss/current_price)-1, to recognize the high risk of AGQ during JL 18 and avoid negative numbers, so now my ratio-to-optimize is just total_gain/average_daily_risk, although there still seems to be something wrong with the math here. Let’s see what results we can get for various tickers and periods:
|2003‥2011||2012‥2013||JL 11‥JL 18||AL 17‥MA 16|
In this table, 2003‥2011 is the back-test period over which the gain/risk ratio was optimized; 2012‥2013 was the forward-test period to see if the system does well with non-training data; in 2014, JL 11‥JL 18 was a problem week of interest, while AL 17‥MA 16 was a problem month.
URTY: I was able to reduce average daily risk while increasing total gain for the back-test period. Gains are not too bad for the forward-test period, but slightly better for July and moderately better for April/May. Risk is indeed lower in every case, but I don’t know yet whether that will actually help me feel better about my trading.
ATML: This was the original ticker I used when constructing the TRIX-based trading model. The new ratio-to-optimize has no effect; there don’t seem to be any parameter values that are better than what I already have.
FNSR: I am no longer trading this ticker because it hasn’t been doing well. I was able to retune the model to eliminate ⅔ of the risk — by eliminating ½ of the profit! This improves the gain/risk ratio, so it’s all good, right? No, that can’t be right. My ratio-to-optimize thingy is for shit. But forward-test results are improved, so … should I use the new parameters or not? I just can’t tell, so this experiment is a failure.
Conclusion: install the new parameters for URTY, but the system still needs work and I don’t know how to proceed.
|Parameters for URTY|
|trixQ||27, 2||28, 2|
|trixS||97, 2||98, 30|
|chandelier+||312, 11.3||312, 11.3|
|chandelier-||28, 9.6||22, 6.0|
|retest||999, 1.0||999, 1.0|
|Buy date||Buy price||Sell date||Sell price||Acct Profit|
|FRED||JL 11 09:30||(Skipped)||$14.96||AU 29||$14.53||-0.41%||+0.00%|
|YHOO||JL 22 09:30||$33.48||$33.56||(Not yet)|
|URTY||AU 25 12:00||$87.01||$87.11||(Not yet)|
FRED: Fell out of the bottom of its symmetric triangle and hit my stop. As things turned out, it’s good that I didn’t buy it!
URTY: The new parameters had no effect on the buy-date for this trade; we’ll have to see if there’ll be an effect on the sell-date.
UWM: There were “Buy” signals every day this week, but the limit-prices were never reached. Try again for next Tuesday.ATML: “Buy” signal for next Tuesday.