Finding an Edge

[padutrader]

The only true edge in trading is utilization of prudent risk management. There is no other edge period.

you need a balanced strategy.....nothing will stand on one leg....without a trend-i call a two bar move in the same direction, a trend- you cannot trade.
without risk management or excellent market knowledge or a healthy dose of patience-which i lack- you cannot survive

the main point is that whatever happens in a market is a process and is usually a slow process because distribution / accumulation takes place at prices beneficial to the manipulator.

That is why market directional moves are sharp and sudden because the market will not be allowed to move from prices beneficial to smart money until the process of accumulation/distribution is finished
this process if understood can be spotted and trades may be made along with the manipulator/smart money.
this understanding is the only true edge and risk management is only an exit when things go wrong and badly wrong,equivalent to a fire or earth quake in real life.

risk management is no solution and cannot compensate for lack of understanding of market movements

 

[padutrader]

most important concept in scalp trading, to me, is padu's claim, "ALSO BE AWARE THAT THE MARKET DOES NOT TURN ON A DIME. I do not beware that priciple, I embrace it.

counter trend traders have to beware as do novice traders who think there are one bar reversals.
yes scalping is actually trend trading but the small trends also do for scalping

 

[tomorton]

The only true edge in trading is utilization of prudent risk management. There is no other edge period.

Its AN edge, its not the only edge.

The situation in which it would be the ONLY edge would be where the trader makes random entries, e.g. in blackjack where the payer is dealt a random hand and "enters" the game without seeing the cards, but can adjust their risk management as the play progresses. Surely you're not suggesting all entries are random, nor that we should all start taking random entries?

 

[Simples]

Black box, gray box, white... way past me...quick explain pls.

Back in the days, you had white box testing and black box testing. White box meaning you have access to sourcecode, so direct access to more invasive tests. Nowadays, this is usually covered in code review and unit tests. These tests make assumptions of what's "inside".

For Black box testing, you had a binary blob, or at least not access to sourcecode. So you make tests and test with less assumptions about internals. Such testing get less invasive, but maybe more expansive and creative. Nowadays, this may be covered in integration and acceptance tests. Instead of binaries, you're probably accessing a service API of sorts.

For other areas, white box would mean manually handcrafting the logic and maths. Black box is usually ML. The computer making the program for you, based on your optimization criterias.

Not sure grey box is well-defined, but could be a sort of mix between the two approaches. For black box/ML you shouldn't assume anything about the internal features, but for grey box you probably can for known areas. Could be a mix, or could be specifically designed for inspection.

There's certain benefits of being able to inspect, but pitfalls when breaking encapsulation in order to curve fit. Future ML applications will most probably be more and more grey box instead of purely black boxes. Them becoming white box would mean internals to be fully human readable, though this is paradoxial since NNs are best suited for tasks tough to manually program. Though, would be better in order to avoid "Plug & Pray".

 

[TraderEX]

Back in the days, you had white box testing and black box testing. White box meaning you have access to sourcecode, so direct access to more invasive tests. Nowadays, this is usually covered in code review and unit tests. These tests make assumptions of what's "inside".

For Black box testing, you had a binary blob, or at least not access to sourcecode. So you make tests and test with less assumptions about internals. Such testing get less invasive, but maybe more expansive and creative. Nowadays, this may be covered in integration and acceptance tests. Instead of binaries, you're probably accessing a service API of sorts.

For other areas, white box would mean manually handcrafting the logic and maths. Black box is usually ML. The computer making the program for you, based on your optimization criterias.

Not sure grey box is well-defined, but could be a sort of mix between the two approaches. For black box/ML you shouldn't assume anything about the internal features, but for grey box you probably can for known areas. Could be a mix, or could be specifically designed for inspection.

There's certain benefits of being able to inspect, but pitfalls when breaking encapsulation in order to curve fit. Future ML applications will most probably be more and more grey box instead of purely black boxes. Them becoming white box would mean internals to be fully human readable, though this is paradoxial since NNs are best suited for tasks tough to manually program. Though, would be better in order to avoid "Plug & Pray".

Whoa! Thanks for that and appreciate the time donation. Don't know much about it. Plug and pray, that's a truism.

 

[tommcginnis]

I’ll try to expand. For a black box strategy, it would indeed be easy,...

Personally, I’ve found gray box strategies provide the most robust solution....A white box can work, with more work, but can lead you down a slippery slope...
That said, I hope you realize how difficult it is to run efficient walk-forward analysis with the latter approaches, as they require periodical parameter adjustments.
Add in context, news, AI, none of which I can, or want to manage, and it becomes obvious that a team effort is required.
I work very hard to keep up with my gray box strategies, but my analysis is very multi-layered and expansive, with the idea that I never end up with a dry well. Not fun, or exciting anymore. Just routine and boring.

Not at all sure what you're getting at here, but I think you're creating new meanings for black-box (unknown content where inputs give outputs), versus plain/visible. White? Gray?
Dunno.

Nonetheless, and as I wrote above, a from-the-start walk-forward analysis of a rules-based system, on a year's chart with 1-day candles, takes me ~2-3 hours, with noting when the rules trigger entry/exit trades, marking down market prices off of the candles, adding up the individual trades, and doing some rudimentary performance. To just eyeball things might be a 5-10 minute operation. To load this into a spreadsheet (market data + indicator templates) would also take 5-10 minutes, once the spreadsheet is formed.

This is a thread on finding/keeping an edge.
The idea that any of this presents an onerous workload?? I don't get it.
The idea that any of this is static? Fixed in place, without reflecting the realities of the dynamic market in front of it? I don't get that either.

To get back to Spectre2007 terms, down the road, I would expect that a nice system is not only hard-coded, but also self-evaluated, re-scored, *modified*, and the changes tracked, such that any paths on those parameter changes can be readily discerned over time. That would be sweet, as well as common sense. Once you find an edge? Sharpen sharpen sharpen.

 

[digitalnomad]

@tommcginnis

I explained it the best I can. @Simples did a good job of explaining the box structures, but getting back to the point of the thread, you won't find a market timing edge that beats the index benchmark using MA crossover trading rules on 1-day candles. You need to get much more adaptive, formulaic, frequent, and granular than that, which also increases testing difficulty. No free lunch

 

[Buy1Sell2]

risk management is no solution and cannot compensate for lack of understanding of market movements

Actually, it's the only solution in trading. There is no other. --Doesn't matter whether entries are random or not. To say differently, shows that you don't have a full grasp on prudent risk management. - - - It's not just about cutting losses.

 

[digitalnomad]

Back in the days, you had white box testing and black box testing. White box meaning you have access to source code, so direct access to more invasive tests. Nowadays, this is usually covered in code review and unit tests. These tests make assumptions of what's "inside".

Nicely explained post above from the perspective of end user. An algorithmic trader should always have full access (white box) to the source code, whether they engineered the strategy, or not. Assuming that blindly running black box strategies is a pipe dream (which I believe), the progression stages for a solid strategy design would involve starting from scratch with a white box, followed by jumping back and forth from white to grey box. This can continue for quite some time, and still not provide pay dirt, but the goal is to get out of the white box ASAP, or you'll be chasing your tail until you're six feet under. You need to be pretty sharp to create anything worthwhile. Most retail traders remain stuck in a perpetual white box, jumping around from one idea to the next, or go broke before allowed the opportunity to figure it out.

 

[tommcginnis]

I getcha now.

Nicely explained post above from the perspective of end user. An algorithmic trader should always have full access (white box) to the source code, whether they engineered the strategy, or not.

Agreed.

Assuming that blindly running black box strategies is a pipe dream (which I believe), the progression stages for a solid strategy design would involve starting from scratch with a white box, followed by jumping back and forth from white to grey box.

Completely understand. And, "Agreed."

This can continue for quite some time, and still not provide pay dirt, but the goal is to get out of the white box ASAP, or you'll be chasing your tail until you're six feet under. You need to be pretty sharp to create anything worthwhile. Most retail traders remain stuck in a perpetual white box, jumping around from one idea to the next, or go broke before allowed the opportunity to figure it out.

Here I would disagree, though -- the goal is not to get out of the white box, but to make the white box produce. The 'jumping around from one idea to the next' is part-and-parcel: "Always be learning." --- which I think, brings us back to Spectre2007s post.

(( This would all be so much simpler with a beer and a white board, no? ))