Support Board
Date/Time: Tue, 26 Nov 2024 09:39:43 +0000
[User Discussion] - Replay is skipping days?
View Count: 1390
| [2023-12-04 02:22:23] |
| j4ytr4der_ - Posts: 938 |
|
I've been doing a lot of replays lately, sometimes over quite long periods of time like 1 year. I run these at somewhat high speeds, but not crazy... usually 480x. Session times are Chicago time day session, with the evening session set to be 10 minutes long (08:20:00 to 08:29:59) to minimize time spent sitting doing nothing as it trades only during the day session. Today something in the TAL caught my eye, and I looked more closely using the Period Stats set to Daily. I noticed that scattered throughout a year of replay, there are many missing days. These are perfectly normal, valid trading days. Not holidays, etc. But for reasons unknown to me, they've been completely skipped over. I can go back and run just these days individually, and they perform perfectly fine. But they've been skipped from the original replay run. I've now looked more closely at two different year-long runs I've done and both are missing various days throughout. Any ideas on why this would happen and how to prevent it? I'm fairly confident that running at low speeds would solve this issue, but that's not really practical when one wants to test a relatively long sample of time. |
| [2023-12-04 16:39:08] |
| j4ytr4der_ - Posts: 938 |
|
Oh also I should've mentioned, I'm using Accurate System mode, tick data.
Date Time Of Last Edit: 2023-12-04 17:48:05
|
| [2023-12-04 21:42:58] |
| User61168 - Posts: 403 |
|
Hi J4y, You probably know this already... troubleshooting such issues is tricky if 1) you are unable to consistently reproduce the issue 2) share every bit of what and how you are testing your algo i.e. chart type, interval, stop&reverse, adding to profits, entries on bar close etc etc. In my 4+ yrs of doing market replays (with simple alerts only), I have come to the conclusion that algo does some stupid shit on it's own will and if you add replay speed into the mix, it is next to impossible to pinpoint the exact issue. I can recall 100s of occasions when simply rerunning the algo at a slower speed or running in 'calculate at every tick' replay mode would generally come to rescue. My overall outcome of market replay functionality is to never trust the exact outcome unless running in 'calculate at every tick'. All the other 3 options are just to 'unit test' the logic and functionality. I have seen numerous +ve expectancy strategies go -ve for non-time based charts with entries/exits on candle close with stop/reverse logic. Even outcome varies at different speed settings. The best I have found is Calculate same as real time option with speed under 240 to be more reliable than others. Obviously, all this depends on your strategy. With that said, some suggestions purely based on my own personal experience : 1) If you are repeatedly using a specific Sim account, better to do a manual clear trade log first before starting the replay. Don't ask my why as I am convinced market replay does not accurately clear the entire trade activity log. Random issues happen in such inconsistent manner that there is no need to bother support with such issues and waste their time. 2) try to avoid using a start date/time in replay mode. Move the chart to your chosen first bar and then do the start. Apparently, this is also recommended by SC. 3) Chart Settings > Date Limiting start date is what you would want to match with your replay start date. Always use new bar at session start option under session times, move to the first bar on the chart and then start the replay. 4) Replay first day of your backtesting duration at slower speed and then gradually increase the speed (using the mouse wheel) to the point when chart starts to skips printing bars. I know this is painful but I have found that running a replay when chart is actually moving to be more accurate and reliable. 5) Suck it up, accept the anomaly and restart the replay instead of spending hours troubleshooting your algo or to figure out why or what. 6) Do not connect to data feed when running replay. Disconnect the feed. 7) Do a max Zoom in to print only one bar on the entire chart. this will speed up the replay and results will be more accurate 8) Use "Hide All" command button to hide all the damn studies during replay when you are not watching. All disable studies that have no meaningful impact to your automated trading. Most visual aid type studies have no reason to be enabled. 9) Avoid changing replay mode while the replay is running. I will end here with a disclaimer that I have no substantiated evidence of what I am saying here is 100% accurate. These are just my own experiences and things I do so please take it at face value. BTW: Today something in the TAL caught my eye what on earth is "TAL" ??Hope this helps. |
| [2023-12-04 22:08:45] |
| j4ytr4der_ - Posts: 938 |
|
TAL = Trade Activity Log. =) Thanks for the reply, appreciate your thoughts. There are many problems I encounter with replay in fact, and they are very reproducible, but typically not "on demand". I can run the same replay multiple times and for sure they will happen, but not always at the same place at the same time, etc. So it's all around quite difficult to troubleshoot what's going on. These issues happen regardless of what I'm testing, from a brainless moving average crossover, to more involved order flow analysis... the same issues crop up again and again. Now to your list of suggestions, I'll address them individually. 1. I always manually clear so no issues there. 2. I definitely use manual dates because otherwise it's utterly impractical/unrealistic to do anything. If I want to replay 2019, I can't literally load 4 years of data and then scroll back to the first trading day of 2019 to start. I'm usually using start & end times in fact, and have gotten extremely comfortable with how they work including some of the quirks of using date ranges. 3. Can't match the start date with the start of the replay, if using studies which would require data to be loaded from before the start time, such as for example ATR, a moving average, etc. ANYTHING that uses historical bar data, even recent range highs/lows and so forth. 4. Yeah also not realistic unfortunately, as much of the time I am running these replays overnight so that in the morning I can begin to look through the data. I'm not literally sitting here at the replay, unless I'm actively working on the logic of what I'm automating etc. 5. Oh this isn't an issue with anything I've built, I'm quite confident in that. So I'm not troubleshooting anything I've done... I'm troubleshooting SC itself at this point. And re-running the missing days is what I've been doing, but to do that I have to first even figure out that there ARE missing days, which means going day by day through the list and looking for a day with no trades - and that only works on a fast intraday system that I know would trade every day. If it's a slower system that genuinely might not trade any given day... then this is an impossible task unfortunately. 6. Yep I always disconnect unless I actively need live data. In fact I wrote a study to ensure this. 7. Aha clever trick, I like it. I tend to minimize the drawings on my charts anyway but this is taking it a bit further, will look into this for sure. 8. I didn't think just hiding drawings did a whole lot unless you had tons of them going back enormous amounts of time, since they are still calculated when hidden? But I'll check this out and compare, and see if I can detect any noticeable speed improvement. 9. Yeah you can't change replay modes while running anyway... I mean you can change the dropdown, but it does absolutely nothing, in my experience. It only actually changes when you stop & start again. Though I never, ever change the replay type during a replay anyway. As for relying on results... that is something I would definitely never, ever do. I don't care how great the backtests are, all that does is show me there might be something to the idea I'm exploring. Nothing ever really counts until it's forward, in live markets. Thanks again... |
| [2023-12-04 23:11:24] |
| User61168 - Posts: 403 |
|
but not always at the same place at the same time - this is exactly what I meant. It used to drive me insane. Now, it is all business as usual to me lolI will respond to 2,3,4 just to avoid any misunderstanding. #2 and #3 - Please read a recent update made my SC Engineering to their documentation regarding use of start date/time checkbox. You can use the "home" hotkey to take you to the beginning of chart and include whatever historical days your algo needs.. then place the chart so that last bar displayed is where you want to start your replay (match this with your session start time) and then uncheck the "use Start date/time" checkbox before clicking start. It takes a few seconds to do this. #4 - you misunderstood what I am saying. Do what I suggested only on the onset of start of replay (for first few minutes) to understand the relationship between your computer's CPU/GPU clock speed etc on how fast you can run a replay with chart moving rapidly. Depending on chart update interval setting, this is a variant and unknown until you do this once for the strategy to come to a realistic replay speed. just one time per strategy and not for every replay re-runs. As for relying on results... that is something I would definitely never, ever do. I don't care how great the backtests are, all that does is show me there might be something to the idea I'm exploring. Nothing ever really counts until it's forward, in live markets. I dont even bother with forward testing unless I have run atleast once in calculate on every tick accuracy mode. Good luck fighting your daemons with replay issues. The worst one is a random trigger of dozens of contracts per trade in the middle of a replay lmao. #10 - I have one more. Date-exclude all partial trading sessions throughout the year. I do this once in december for cover the next year. I also exclude public holidays and usually a day before the holiday if it falls on a monday or friday. To address your dilemma on how to find missing days in TAL, here's what I do in Excel. Step1: Cell A2 enter "Period" Cell A3 enter 1/3/2023 (first trading day of the year) Cell A4 = =IF(OR(WEEKDAY(A3)=1,WEEKDAY(A3)=2,WEEKDAY(A3)=3,WEEKDAY(A3)=4,WEEKDAY(A3)=5),A3+1,IF(WEEKDAY(A3)=6,A3+3,IF(WEEKDAY(A3)=7,A3+2,0))) Drag this formula to next 260+ rows to give you all the weekdays Step2: delete rows when CME holidays. This will give you your actual trading days throughout the year (Step1 and 2 is required only once to save it in excel for repeat use) Step3: After each replay run, copy all from Period tab and paste it in Column B to quickly determine if there are any missing days :-) Protip: All this should be part of your overall equity curve analysis. Attaching a screenshot since I am in the mood to share today. Date Time Of Last Edit: 2023-12-04 23:19:06
|
 equity curve.png / V - Attached On 2023-12-04 23:11:17 UTC - Size: 162.97 KB - 62 views |
| [2023-12-04 23:18:30] |
| User61168 - Posts: 403 |
|
Some more tips I shared in the past here >>> speed backtesting raffect reliability and this thread has the suggestion from engineering on not to use the start date/time.... Issues to replay on a new Intraday Chart Happy reading! |
| [2023-12-04 23:19:48] |
| j4ytr4der_ - Posts: 938 |
|
Thanks for the additional followup. I'll check the docs again, I've read them before but maybe there's been a change. I don't really follow what you've written but maybe the docs will make more sense to me. As for #4... ok I understand what you mean now. I've already done this actually, and what I've found is that yes, of course there is a speed slow enough that everything is totally solid and reliable. The problem is, for example with something I'm testing currently that speed is 60x. This will take an enormously long time to work through a year of data. Even running 6 instances with each doing 2 months (as I often do), this will take much longer than it should. It's just so disappointing to not be able to just let the software run "as fast as it can" and not worry about such things. As for the Spreadsheet suggestion, unfortunately that won't help when you have a strategy that doesn't always trade every day. In particular when I'm working on order flow ideas, they often legitimately don't get a trade in a particular day, so the only way I can know if that was a valid no trade, or a replay glitch, is to go through the list, find all the days that didn't trade, then look at the chart of that day. I understand that this is a way to find only the weekdays, there are other ways to do that as well so that's not a problem. The trouble is that I can't rely on SC to trade every time, any day, at any speed, when there is a valid trade to be taken. I was pretty sure I could fine-tune things to "just work" but I'm less confident in that assessment now. Thanks again, just saw your additional response so I'll check out those links. |
| [2023-12-04 23:21:50] |
| j4ytr4der_ - Posts: 938 |
|
Ah I see now, I recall that thread... you only do 30x speed, so right there that's the answer. I'm not willing to accept that limitation, I already can do 60x and it's fine but that's absolutely not acceptable to me. I'm going to just have to figure something else out, probably just more instances to run in parallel is my guess. Definitely some other solid tips in that thread as well so appreciate you posting it. |
| [2023-12-04 23:29:27] |
| User61168 - Posts: 403 |
|
You are welcome. you have a strategy that doesn't always trade every day - I am sure you have thought of this already. To tackle this, put all your entry signals in a single color bar alert condition with "sum of alerts" subgraph enabled. This will return a total number of trades signal that you can then match up quickly with total trades in the statistics tab after the replay ends :-
Date Time Of Last Edit: 2023-12-04 23:30:46
|
| [2023-12-05 00:27:45] |
| j4ytr4der_ - Posts: 938 |
|
Yeah, great idea... except that many times I'm testing a system where the signal no longer exists when the bar closes. But yeah, on any system where it will always be valid at bar close then this is a great solution at least to telling you if you need to look more closely, or not.
|
| [2023-12-05 13:27:23] |
| j4ytr4der_ - Posts: 938 |
|
Frankly at this point, I'm leaning towards just only ever using tick-by-tick for replay. I was using Accurate System as from my reading of the docs, it looked like it was identical to tick-by-tick (unless I'm using orderflow, in which case there's really no choice but to use tick mode) so was hoping to gain a bit of speed that way. But it's so riddled with issues that I think the only feasible speed advantage is going to have to come from parallelism.
|
| [2023-12-05 15:03:26] |
| j4ytr4der_ - Posts: 938 |
|
Just to follow up, I've done some comparisons this morning for speed. Have come to a few conclusions (running at 30720x for these tests to just let it run "as fast as possible"): 1. Calculate At Every Tick/Trade is the only mode that can be relied on (to the extent that any mode can be relied on). Every other mode is guaranteed to miss things, skip trades, skip days, etc. if you're running at anything resembling a high speed (like above 60x). 2. Zooming in to see only the current bar, is making no difference in terms of replay speed. 3. Hiding all studies definitely does help. Not massively, but a measurable speed boost. So at this moment my final conclusion is that maximum efficiency, consistent replay will only be possible by using Calculate At Every Tick/Trade, hiding all studies, and running many instances in parallel (and off a RAM disk for truly squeezing every possible bit of performance out of your machine). Beyond that I don't think there's much else that can be done to accelerate anything further. Date Time Of Last Edit: 2023-12-05 15:19:56
|
| [2023-12-05 17:34:27] |
| User61168 - Posts: 403 |
|
Couple of more suggestions: 1) Move this to user discussion to get additional perspectives from other advanced users in the field of performance engineering 2) don’t forget to disable Trade activity log during replays. 3) hide/destroy or close all other charts and leave only the replay chart visible 3.1) try to hide the replay chart while running replay to see if it speeds up 4) disable message log also during replays and get rid of all the damn alerts that gets written in the alert windows to minimize disk I/O 5) Just yesterday SC confirmed it is okay to use as many installations of SC on as many computers as we want as long as we do not connect the data feed. This will help you in running parallel. Have a clean copy of install and then copy the scid data files across the x number of installations 6) if you plan to indeed go over this architecture then in my opinion (but I could be dead wrong on this) it would be better to just avoid using continuous contract option and split the data files into each historical contract scid files. I will admit I don’t fully have a grasp of all the intricacies involved with continuous contract settings and if the desire is to eliminate doubts, this might be another way to do the divide and conquer. With all that I have written about replays, I am slowly moving to a conclusion that massive amounts of backtesting is a waste of time. Purely assessing my own journey, I have had no success whatsoever in finding a single strategy that works across more than 12-15 months of historical data (without considerable drawdowns). Almost all have shown decay in performance from one year to the next and I just have not been able to decipher whether they are a cause of anomalies in the replay architecture or something else like markets do change year over year. I have also remained strictly focused on NQ since 2020 but I hesitate to share this experience and be the source of discouragement for others. Anyways, just a massive area of research my journey has been so far. I don’t think there exists a algo that could run year over year without tweaks but I do hope I am wrong :-) Date Time Of Last Edit: 2023-12-05 17:38:33
|
| [2023-12-05 17:48:45] |
| j4ytr4der_ - Posts: 938 |
|
Yeah SCE already made it a user discussion so... maybe it will get more useful stuff over time. Good points about the logs, I actually normally do have them all closed during replay and have only had them up recently while troubleshooting these issues. Definitely for raw speed though, closing them all can make a massive difference. I haven't been disabling the message log though, good idea there. As for your experience... actually I think more people need to hear exactly that kind of thing. Automated trading is (in my opinion anyway) not something that will ever be purely "set & forget" where you have a "strategy" you can just run forever. I see it as a constant "job" of identifying potential edges, validating them, deploying them small to prove out the validation (backtest) in live markets, then running them for as long as they're working, while you find the next edge(s). Treat them like a portfolio of sorts, always have a dozen or so at your disposal, and continuously drop the poorest performer and replace with something new/better. Even better if when an edge stops working you can define what changed... because maybe now you know the right conditions for that edge. I think too many people see automated trading as some magic cure to their trading ills, and so they do want to give up when like you said, they can't find anything that just works forever in backtest. I don't think there is any such thing, so I'm not even looking for it. A variety of real edges, in a variety of markets, trading small, taking minimal risk and distributing it. That's my aim anyway. =) Good conversation, appreciate your input! |
| [2023-12-05 18:03:38] |
| Sawtooth - Posts: 4120 |
|
With all that I have written about replays, I am slowly moving to a conclusion that massive amounts of backtesting is a waste of time. Purely assessing my own journey, I have had no success whatsoever in finding a single strategy that works across more than 12-15 months of historical data (without considerable drawdowns). Almost all have shown decay in performance from one year to the next and I just have not been able to decipher whether they are a cause of anomalies in the replay architecture or something else like markets do change year over year. I have also remained strictly focused on NQ since 2020 but I hesitate to share this experience and be the source of discouragement for others. Anyways, just a massive area of research my journey has been so far. I don’t think there exists a algo that could run year over year without tweaks but I do hope I am wrong :-)
I concur. This is also my experience.There is no such thing as a set-and-forget autotrader that will perpetually be profitable. I have built hundreds of autotraders over the years and none were profitable for more than a few months. Some were very profitable for weeks, then gave it all back and more. Every autotrader requires periodic tweaking (or abandonment), so that means all are actually semi-automatic. A semi-automatic system could give you an edge, but nothing compares to wise discretion. |
| [2023-12-05 18:05:47] |
| User61168 - Posts: 403 |
|
Thanks. Journey continues as this is a fascinating research project for me to understand how markets (maker) operate. It’s an interesting challenge to find a universal market agnostic strategy which works without giving any regard to trending or ranging markets, without using any indicators or order flow or looking into the past historical data. I think I will need AI/ML and petaflop compute power lol
|
| [2023-12-05 18:15:42] |
| User61168 - Posts: 403 |
|
Thanks for the validation Tom. It means a lot coming from you. The urge towards Pursuit of financial freedom is what keeps me going. Continuous tweaking is not something I am willing to accept as a long term strategy :-)
|
| [2023-12-05 19:27:53] |
| Sawtooth - Posts: 4120 |
|
@User61168, @j4ytr4der_: Thanks for the great tips on backtesting! See attached for a matrix of my opinion of various features/limitations of the backtesting choices. I've hesitated posting this in the past because it could become outdated over time. In fact, it's not comprehensive as is. |
 Backtesting matrix.xlsx - Attached On 2023-12-05 19:27:36 UTC - Size: 12.87 KB - 191 views |
| [2023-12-05 19:29:30] |
| User61168 - Posts: 403 |
|
consistent replay will only be possible by using Calculate At Every Tick/Trade, hiding all studies, and running many instances in parallel (and off a RAM disk for truly squeezing every possible bit of performance out of your machine). Beyond that I don't think there's much else that can be done to accelerate anything further.
Another idea you could incorporate to speed up your repeat replays: Once you have a full run, go to trade activity tab, copy/paste the entire log into excel. then apply filters on your entry and exit study names to generate the list of days when your algo traded. Next step is the determine the days when no trades are executed. Once you have the non-trading days, the use the fantastic date-exclude feature in sierra charts via global symbol settings > go to symbol and look for custom date exclude field to cut and paste all your non-trading days. Then do a chart reload and recalculate for setting to go into effect. After this step, you will see all the days where you have your trade fills for quick analysis and for future replay re-runs by drastically cutting your replay time. I do this almost everyday with every re-run to analyze all the losing days. It brings huge benefit that no other trading platform I know of is capable of doing. Attaching Screenshot that shows how to generate the exclude string in excel via simple cut/paste. |
| exclude days.png / V - Attached On 2023-12-05 19:28:03 UTC - Size: 81.45 KB - 60 views |
| [2023-12-05 20:10:55] |
| User61168 - Posts: 403 |
|
And Lastly, I will say this..... thanks to Sierra Charts, I have come to the realization NOT to trust any strategy (from system sellers) that claim to have positive expectancy. It's the biggest lie in the trading industry. Give me any failed or successful strategy and I can convert it into a positive expectancy strategy in less than 30 mins. It's very simple and straight-forward! Here's how: 1) Look at any trade activity log results 2) select either day of week or time of day filter 3) pick the winning days or hours 4) apply time of day filter or day of week filter 5) Now you have a nice looking equity curve showing positive expectancy. It's probably the reason why I also do not believe in "statistical" edge as exporting chart data in excel and picking a "setup" based on historical data is also very easy to do. With that said, maybe there is a way to auto re-invent strategy recursively based on last week/month trading stats i.e. self-learning AGI algo just like humans learning from our past experiences and failed attempts :-))) |
| [2023-12-05 22:24:48] |
| j4ytr4der_ - Posts: 938 |
|
You're right that it is ridiculously simple to generate a good looking equity curve - this is trivial. But I most definitely don't agree that there are no "statistical edges". There absolutely are, but they don't come from a trader's *results*. That's not statistics, that's "resulting" - judging the quality of an idea by its results. This is a totally invalid approach. A real statistical edge is based around things that happen in the markets with a repeatable, statistical frequency, that you can exploit for positive expectancy. This is absolutely real but it's not very sexy, and I don't think very many people think this way and even fewer apply it to their trading. |
| [2023-12-06 11:20:56] |
| User61168 - Posts: 403 |
|
I most definitely don't agree that there are no "statistical edges" I agree that there are statistical edges but I don't believe in them for reasons we have already discussed i.e. they don't work consistently and have a limited shelf life. Some questions that run wild in my mind when I think about this: 1) If a pattern or mathematical relationship or market structure repeats based on past historical context, please help me understand why there is such a high rate of failure across retail algo traders. Every algo based on a pinbar or a 123 pattern or whatever should make millions based on how appealing they look when we scroll the chart backwards :- 2) What is the root cause of algos to stop working after a while if they are based on such proven statistical edges? 3) repeatable, statistical frequency How long and how often should an edge have a repeatable and statistical frequency for it to be considered a robust edge? I would love to see some specifics based on your real live experience and not based on garbage from youtubers and influencers. I mean what all trading statistical measures and benchmarks proves that an edge will atleast meet the historical results in the future. None imo but I could be dead wrong.At a holistic level, what gives us confidence that a strategy would succeed? Every trade statistic out there is based on some historical context or recurrence and eventually prone to failure. It's the fundamental and structural flaw in the methodology we all use but that's all we have available to us retail crowd in regards to technological advancements. Comparison with Citadel/Rentek and institutional/prop firms do not count as they have millions to spend on compute infrastructure and army of PhDs. Date Time Of Last Edit: 2023-12-06 11:31:17
|
| [2023-12-06 13:53:22] |
| j4ytr4der_ - Posts: 938 |
|
Let's examine some of this =) 1. The patterns you describe are meaningless and arbitrary. They're not based on the actual volume of what went on in the market, so they are essentially just random noise. Any price pattern traded mechanically, will always ultimately lose money. There are some traders who manually manage such trades and they're really good at getting out once proven wrong, so they can keep losses small enough to make money over time. But that just means their edge is in keeping losses extremely small. 2. It's not the algo that stops "working". It was the underlying premise in the first place that never made any sense (assuming they're based on things like you describe). Markets also change, they go through phases, cycles, etc. So any expectation of any mechanical solution winning 100% of the time is totally foolish. 3. I don't think this can be answered since "robust" will be a subjective measure. But if an edge has say an 82% frequency of occurrence, and one applies expectancy and position sizing rules to it, then it should "work" if not forever, then at least for quite a long time or reliably under certain market conditions. I don't have anything personal to share at the moment as I'm in the early stages of actual deployment of my "bots" and am still discovering mistakes in what I've done, etc. So far I've passed evaluations with them (my usual testing grounds), and lost some cash (due to aforementioned mistakes in what I built). But I can see the principles are sound, and I'm still on a much better path than ever before in over 20 years of "chasing the trading dream". This idea of some mechanical black box just printing money forever, is what I mentioned as a pointless thing to chase in an earlier post. Took me a long time to let go of that. Markets are forever changing, and everything has to adapt. This is why my current desired approach is to have a number of such edges, and constantly be working on more, so that I am regularly reviewing performance, testing new ideas, and replacing the poorest performer(s) with new ones, in the same way any "diversified investor" would replace their poorest performing assets in a portfolio. My job is R&D, the computer's job is to trade. As to your last paragraph... I would say it's a myth that things like phd level math and an army of quants are required for this sort of thing. In fact basic math and some (ok, a lot of) effort are enough, but one has to not get distracted by all the noise out there (particularly in the retail space) and chase nonsense like TA indicators and such. You might like these two videos... they're by Lee Harris of emojiTrading and Trading Research Group, very serious guy, not an "influencer" by any stretch of the imagination. He's active on this board, maybe he'll have something to say about all this. Or maybe not... he probably can't be bothered lol https://www.youtube.com/watch?v=XRzBYwWbAmQ https://www.youtube.com/watch?v=JR3tW7k779I&t |
| [2023-12-06 15:29:41] |
| Sawtooth - Posts: 4120 |
|
Some very brief responses: 1) If a pattern or mathematical relationship or market structure repeats based on past historical context, please help me understand why there is such a high rate of failure across retail algo traders. Every algo based on a pinbar or a 123 pattern or whatever should make millions based on how appealing they look when we scroll the chart backwards :-
Your eye only sees the good entries when the conditions are met, but the algo sees all occurrences when the conditions are met.2) What is the root cause of algos to stop working after a while if they are based on such proven statistical edges?
Because the market is dynamic and the algo is rigid.3) How long and how often should an edge have a repeatable and statistical frequency for it to be considered a robust edge?
A repeatable edge could end at any time.At a holistic level, what gives us confidence that a strategy would succeed?
An autotrading strategy will always eventually fail, so the confidence of success is to use its signals with discretion, not with automation.
|
To post a message in this thread, you need to log in with your Sierra Chart account: