True. But you are also gambling if you assume that your mechanical backtesting results are accurately reflective of the future state of the market. Trading is gambling by its nature.

The trick is whether you're able to consistently keep losses small and winners large. If so, you could trade using a Magic 8 Ball as far as I'm concerned. It could be argued that all these trading systems, discretionary or mechanical, are really heuristics designed to override our natural tendency to do the opposite: exit winners early, and hold onto losers in the hope that they'll "turn around." Use whatever method gives you enough confidence to override those natural instincts.

The reason I personally favor discretionary trading is that I've perceived that market conditions change more frequently than some traders appear to assume, in a way that can render a formerly profitable system into a dead loser.

If that's the model of the market you're building from, then the question becomes not, "What set of mechanical rules has worked best, on average, over the last X thousand trading days?", but rather, "What's working now?" And if that's they question you're trying to answer, then of course you will develop new rules in response to changed conditions. Hopefully not totally unfounded rules, but quite possibly without enough of a statistical background to run a comprehensive backtest. If you find, say, a correlation between asset classes that appears to have developed recently due to some issue that's at the forefront of the news, then why not trade it until it stops working? The fact that the correlation is new has no bearing on whether it's working at the moment.

Understood, but not my approach. Unless I know the expectancy of a particular setup, I'm not taking it. If I notice a new phenomenon in the market, I go back and see how often that phenomenon happened before, and how profitable it would have been to trade it, otherwise, I consider it a random phenomenon, and I don't trade randomness. If it happened 6 times in the past 3 years, it's not statistically significant enough for me to trade, I still consider it random. It it happens in a market environment I've never seen before, it is outside of my comfort zone. During the last flash crash, I saw dozens of opportunities to enter the market, but I did not, because that is not the market environment I am use to trading.

Auto trading (for most platforms) leans toward market orders. limit orders present numerous "gotcha" coding difficulties. Partial fills. Basically, it opens up several decision tree possibilities that must also be included in your code. (it can be done, but it's very difficult).

Auto trading challenges and difficulties can push you toward more simplistic methods (which I'm not so sure is a bad thing).

In addition to challenges with limit orders (encouraging market orders), it also encourages "next bar" execution, rather than intrabar. There are several challenges with backtesting, then again, these challenges are present whether you trade discretionary or auto.

I understand and appreciate all your comments on mechanical trading vs. discretionary.

One other issue here from my perspective: a lot of my "discretionary" trading decisions are really mechanical, but not in a way I would be able to easily program a computer to recognize. By that I mean, patterns that are easily apparent to the human eye, but hard to distill to an objective set of rules.

Take for instance the concept of the "double bottom." It is easy to look at a chart, see a prominent swing low earlier in a day, and then make note of that level the next time it is approached. If price bounces near that level, you might be on guard of a potential double bottom / hold-of-support situation, and be less inclined to short that bounce even if your mechanical system would normally say to short. There's a good example of that today, around 11950 on YM. Initial swing low established around 9:40 ET, then a subsequent retest around 10:30. Without that prior swing low at the same level, I would have expected the downtrend at 10:30 to go further.

I could define an objective rule to capture this situation, but wouldn't know how to tell a computer when to recognize what level to use as the "double bottom" zone. How many bars back do you look? How do you define the concept of "swing low"? What exceptions do you build in? How many ticks can price overshoot/undershoot and still be considered a valid retest of that level? Etc. It's obvious to my flesh-and-blood little pattern-seeking mind but complicated when you try to break it down into its component parts. My fear is that if I tried to program a computer to recognize what I can see intuitively I'd create some "Frankenstein" that goes around taking completely nonsensical trades.

Dear @mattz,

Can you elaborate this:

What is the difference in this regard between automated and discretionary trading?




Thanks,
Arpad

What if while trading, the trader tells himself that the pattern he is seeing just barely falls outside of acceptable parameters, and the trader immediately recognizes it as such? Would that be breaking the rules?

What rules would you be breaking if the trade isn't triggered because it's outside acceptable parameters?
Sounds like you would still be following your rules if you don't initiate the trade.