I have tested in real life and backtesting multiple indicators (macd, aroon, mesa, simple roc, bbands, psar) on every single time frame with literally every single combination (brute force backtesting every possibility of every indicator setting and just going with the top one) that passes metrics...
THEY HAVE ALL FAILED EVERY ----SINGLE ---------TIME
It's worse than 50/50, or I would have made some money by now. I have ran in circles for at least 36 months, and evaporated at least $1000 in testing ($1 bets x 1000 separate test losses).
The pursuit of the perfect TA strategy feels like a drug addiction now that I must quit.
The HODL people were right, atleast then you don't have bot losses from bad calls. I'm about to abandon the entire concept of TA as a waste of time, and with a considerable amount of real life research to back it up.
HODL + yield farming, or running grid, is the only reasonable conclusion I can come to.
But can you really run with no stop losses? Any form of stop loss is a form of TA, and TA is always a lie.
The only thing that sounds reasonable to me at this point is splitting the bag on 100 pairs, the stop loss is diversification, maybe with rebalance passes.
You’re driving while looking at the rear view mirror. You will only get so much out of it and there’s never 100% certainty.
Jaxson Wood
i have gone so far off the deepend with this i think the chart if a reflection of God and quantum physics.
I'm not sure if the past is even relevant to the future anymore; at all, and not just on the chart. You can put anything on the chart; and if TA is a lie, then momentum is a lie, and physics itself breaks down.
An object at rest may accelerate at any time, an object in motion may stop at any time.
Nicholas Howard
>You’re driving while looking at the rear view mirror. this is a perfect analogy; and any solution to playing TA would also seemingly apply to vehicle navigation where your only camera is pointing backwards and you have no map, and you don't know where the intersections are.
You can measure where the intersections might be; on a freeway, every 2 miles, from looking backwards, but it would be nearly impossible to drive a car this way and stay on the road unless the road was always a straight line and you never had to turn.
The answers that I come up with no in my mind are equivalent to flooding the entire highway with ants, some will flow off the wrong exits, but the bulk should stay in the pipe.
Anthony Williams
do you know how to code or are you just using TV strategies reccomended by pajeets on youtube.
you just need a half decent entries and trailing stoploss/take profit for a profitable bot. with a well tweaked trailing tp/sl you can even make profit on a simple ma cross. protip: leverages and tp/sl is more important than win rate / entries
everybody is driving on the same highway. at least TA have a rear mirror and are trying to drive. hodlers are just curled up in the boot
Jaxon Perry
You need to assume that the biggest market movers are going to follow your TA algorithm to begin with, because retail doesn't move sheeit. And that's the problem with TA, you need to convince a bunch of midwits to believe in some gay pattern or lines in order for it to work (exactly why it's important for twitter influencers to convince their retarded followers how they'll make it, but only if they follow the underwater jew, interest rising wedge on the 5th sabbath timeframe), and even one whale dumping a fuckton of btc on spot can make your meme lines useless. Assuming they use TA at all, they've most likely agreed to use the gayest most obscure meme lines to begin with. That's why I trade macro instead of micro since announcements nearly always have the expected price action. Sell even after a 10-20% on some of the most bullish news and your portfolio just skyrockets.
>(1) good news, unexpected = strong poompa (euphoria) >(2) bad news, expected = poompa (relief, priced in) >(3) good news, expected = week doompa (fomo, buy the rumor sell the news) >(4) bad news, unexpected = doompa (panic)
Don't let the midwits fools you into TA like that ben cowen cuck because it's all psychological, not mathematical.
There is only one TA which still continues to work on a longer time frame, it's EMA but I won't divulge in to the configuration of it and timeframes because it's a closely kept trading secret. It will confirm whether we are in a bear or bull very soon with good certainty and will also indicate where to set a SL/TP.
Joshua Davis
I wrote a completely custom bot with it's own built in backtester, that not only can play signals but grid/market make (pure market making is a separate scam, grid seems legit).
i haven't tested hard take profit / stop losses outside of shittokens, and that didnt' play well from the sheer amount of rugs.
playing purely on indicators fails, and I have metrics to shit filter the 1,000,000 combinations of macd or whatever indicator i'm backtesting. i don't just backtest macd with 12,26,9; i backtest ALL combinations from 3-100. 1,1,1 1,1,2 1,1,3 etc
filter then for profit factor and winrate and profit and all time low/high on the indicator and pick the best one. it fails miserably on everytime frame (default settings also fail miserably).
you can only go so far with the zoom out cope, the 1hour or the 4hour doesn't test well either and it's equivalent to an "always long" strategy at that point anyway. and there's not enough 4hour/daily candles on most of these to test zoomed out that far.
the only thing that I've ever had profitable is a pure grid, with no TA whatsoever. grids are equivalent to tp/sl i guess. but it has the problem of >what happens when the coin 45 degree moons or dumps for 6 days straight grid fails hard, but if you can't predict the reversals reliably there's no point in even comparing. it seems like you just have to account for an immediate dump at all times; which boils down to an ultrawide grid.
Nathan Hill
Op the fact you're back testing is a good step you're taking. Now I hope you're testing at least 3 different types of stratergies.
1 for bull markets 1 for bear markets 1 for oscillating markets where this not much action happening
Then for each of these 3, you need different ones for different time lines, a daily one, weekly, monthly, yearly etc
Don't back test a single strategy over a decade of data. In that decade you should be using multiple strategies depending on the market
Anthony Green
>There is only one TA which still continues to work on a longer time frame, it's EMA that's why I've gone hard on MACD, and it backtests well but never plays out in reality; maybe because re-running backtests and flipping settings too quickly; but it shouldn't really matter if the backtester is rolling over 2000 candles everytime on longer timeframes (also failed)
The only thing that backtests well is equivalent to holding the trade open for 500 bars, at which point; of course it backtests well, it only flipped once or twice during the entire 2000 candle backtest period and the test isn't really reliable.
With backtesting there's a massive problem of biasing towards just holding something or not holding something; because that's seemingly the most profitable.
The backtester will bias towards a 100,100,100 macd settings, because just holding the bag and not trading at all was the most profitable outcome; so I have to artificially punish it or straight filter values out, at which point it just biases towards the max again.
MACD is just MA cross, and I can use EMA with the MACD calculation; but it backtests worse than SMA (just adding more lag).
PSAR seems to be the most reasonable thing at the moment because it has no actual bearing on reality; it's just a timer. still backtests and performs like shit and you can't run psar on anything higher than the 1/5M or it's just equivalent to setting a -90% stop loss on every flip
Matthew Walker
Well, use very long timeframes weekly and monthly to be more accurate on buy ins. Using moving averages on large time frames also seem to be more accurate
But be aware that the weekly chart = months for the move to happen. And monthly chart takes years to happen. But if you could catch/dca under the weekly/monthly average, then you will have months and years of gains
>shittokens there's your problem. shittokens are too random. they're just gambling. put $100 on one you either lose it or make 10k. TA only works on the big coins,
Jace Robinson
>Now I hope you're testing at least 3 different types of stratergies. the idea of brute force backtesting indicator combinations (and re-running it every 5 minutes or whatever); should have had the effect of a "one size fits all" strategy.
I'm already experimenting with making a long/short/flat call myself initially and biasing the bot with that.. with the pure grid - no TA grid bot testing. the only call you make with that is starting bag size; either start full, start empty; or 50/50. obviously if it's gonna dump start empty.
i have also experimented heavily with combinations of indicator timeframes (ie 1m MACD must be long *and* 5/15m MACD must be long); but that seemingly didn't work well either; if the 1m is making unprofitable calls to begin with then why use it? (and dropping it improved performance), but they were all unprofitable in the end; so your stacking shit with shit.
Joshua Hall
i'm playing and testing on CEX I only briefly experimented on shittokens with a PCS bot. so many rugs it wasn't worth it.
i've tried shitfiltering pairs for likely success too. >if the bot performs like shit on the pair, why not brute force the pairs and turn the bot on the best backtests?
TA only works on big coins seems to be accurate; or specifically coins with large market caps / decent volume, but even then, on major coins, I have never come up with anything using TA that performs reliably; all of the calls are bad and it eats losses from not only the coin loosing value; but from making terrible calls.
I have also tried combinations of indicators; and a literal "best out of 20" thing where I would run 10 indicators on 5 timeframes and just go with longs > shorts = long, etc. (that just adds lag, it ultimitely fails too).
there seems to be a common theme here
Oliver Bennett
>maybe because re-running backtests and flipping settings too you're over fitting. stop fucking with the settings too much. no wonder it won't work in forward testing, because what you did is fine tune it so much to fit the chart that's in your test range.
i have not testing very long timeframes, daily; weekly or monthy, from lack of data and just questing it in general; testing those timeframes would take forever to actually play out and the losses are much greater (the equivalent stop loss would be massive on every trade).
how much of that is just cope i wonder? If it takes 3 years to play out, it's going to work great until it doesn't. on the lower timeframes it seems you can atleast find out quickly if it isn't going to work.
TA has failed so many times this is boiling down to a spiritual philosophical question. >Does the past matter? The chart is a reflection of human consciousness; and on longer timeframes you would think more data = good and you can actual measure human thought, which itself is a reflection of the universe or God if your religious.
But the question is; is there random? The entire TA theory assumes fate is pre-determined and there is no random. If there is no predetermination though; then the past has no or little bearing on the future, and TA will never work
Hunter Davis
the over fitting concept was deep in this. I was aware of the overfitting problem, and the thought process is "let's test out MAXIMUM overfitting". Let's tune the indicator EXACTLY to the pair, and then to solve the overfitting problem, re-run the backtest/scan every 5s/15s/1m/5m/whatever. clearly it doesn't work.
the alternative is to just go with baseline default indicator settings; but that doesn't backtest well or playout in reality either.
Blake Thompson
TA is astrology for men.
Joshua Edwards
Indicators built on average values are useless.
Austin Gray
Where did you learn TA from? You may have the wrong ideas on how it works.
Jack Morales
aroon and psar didn't test well either. and psar makes the most sense, it's literally just a timer; not now instead of predicting trends your trying to predict market periods and timing.
Carson Cooper
i read on it constantly but it's been a combination of self taught and being brutally destroyed testing bots using it. Initially it was a mess but being wrecked a thousand times is a powerful teacher on what exactly is going on. I understand the math behind the indicators at this point I think I have a decent understanding matched with significant real world performance testing.
I mean really there's only so many things you can go with. >Moving averages (ma cross) variants >ATH/ATL (aroon / support / resistance levels) >Volatility (bbands, raw volatility) >Timer (psar, not sure if there's others) there aren't really 50 indicators they're all just minor variations in 5 different themes; i've pretty much dug in to every single one that talib provides (and finta); and some of the wonkier ones on traderview; but those ones are usually just tweaks on other indicators too
Samuel Bennett
at a certain point with backtesting it doesn't even matter what the indicators are. all that matters is the calls, and their performance metrics.
a mouse could press a long/short button and generate the calls and the backtester would run it.
Brayden Gonzalez
I would have to agree
My "random lol" trades where I just buy or sell at a whim without looking at anything are actually more successful for me than when I do my due diligence and use TA
