Topic: Wouldn't it be nice... (the LazyWhale algorithm) - page 13. (Read 24717 times)

So on a close the signal will never change its value? What I mean is that if the signal is based on the previous number of bars on average then it may change signals if the average changes enough the past few bars. Usually this is the case where you rely on a larger timeframe to make a signal on a shorter timeframe chart... but if the larger time frame closes (6h) and the signal does not change thereafter at any time then I'd say its not repainting, I'd even try the indicator on other charts to test.

Almost got it right, champ.


12 USD (= 1 BTC) invested in October 2012, i.e. 2 years ago.

Current account:

b&h: 1 BTC. USD account value ~=380 USD at current price.

alg.: 1,400 USD. BTC value ~= 3.7 BTC at current price.

The thing with people who keep buying back their bitcoin's at a lower price is they'll always keep getting more bitcoins but their account value keeps going down lmfao...

So if you added current btc price to your technique I am sure you'd be loosing money like everyone else trading bitcoin.

I'm still trying to figure out what the 'repainting' problem is that could apply here? Genuinely interested in learning something, hope you don't mind my persistence.

Quick Google suggests that 'repainting' is what happens if an indicator that relies on close price input seems to yield a signal, but loses it upon getting the actual input close value. That what you had in mind? If so, not really an issue here, I'm using 6h close and a signal is only counted after a finished close value.

That said, a 'whipsawing' signal is always a possibility of course. Didn't really happen in my backtest, and I suspect that because of the longer trend scale I'm using, it's also less likely to happen than in, say, a method that yields hourly signals, but in principle that could happen.

Not a bad strategy.

Yea thats what Iguessed happened.. but i think it only does this after making a new high right? There were cases where the same thing happened but before a new high was made and didnt trigger a sell signal... maybe you have a holly grail but I think pa is always changing especially since bigger traders are jumping on. But I agree that it makes sense on a bullish mindset howver Ithink something like a 200/50 ma on the daily should also do the trick in that case. I think next run up may shake ppl out by making similar tops but continuing up again without pull backs so it wil break patterns because we had 2 serious bears during this last run up so far.

I didnt mean curve fitting I meant repainting which happens to change signals over time as they hapoen in real time but look like they picked tops and bottoms in hindsight. I saw lots of these from ppl trying to sell bots online for forex strategies but they were based on repainting indicators... just want to make sure this isnt happening for you as it might get your hopes up.. hopefully once a signal is given it sticks.

Double answer:


Come on man give me  a break... I never called it a "secret formula". It's, maybe, just maybe, a semi-clever way I came up with to combine automatic signals to get something like "trend confluence". If it performs well enough, I'll post the details eventually. If it doesn't, I'll do so as well, but hope that interest in the method goes away anyway to the point where I don't have to.




Absolutely, vehemently agree with your words of caution. Especially considering that the algorithm, by design, only yields a very small number of data points. (also, thanks for an interesting story about your bots)

That said, I don't see it inherently as any riskier than, say, putting your money on the output of any other automatized strategy in Bitcoin: the traders that run EMA ACO bots have been doing so for a while... I don't think they forward tested a few years

And, as I tried to argue above, the method *is* crafted around a particular peculiar property of Bitcoin: the ability to go through insane growth spurts at time, and never look back (at least by order of magnitude of price).

If you trade for maximal USD preservation, this method won't do much for you. It's aimed at those who, in principle, believe Bitcoin is quite likely to go through another price era change at some point, but until then, trades with ups and downs. Which is, more or less, my own perspective on it. The goal is then to accept a higher than usual risk in a strategy (e.g. accept a higher drawdown), but limit the tail risk at least to /some/ degree, compared to a pure b&h.

Plus, like I said two or three posts above: the buys are triggered "easier" than the sells, so reigning them in with a stop loss might make sense.

EDIT: I get why you're saying the "sell at double top looks like curve fitting", but I'll say this much: both 2013 and 2014 (and similar in 2011, which i didn't post), the sell of comes after the initial bounce back, when the post peak "consolidation" fails. I think I found a way to encode that structure in a reasonably general way s.t. the parameters fit all previous cycles and at least /stand a chance/ to fit the next one as well (assuming there will be another "bubble"), or otherwise make sure limit the drawdown (assuming there's no further bubble).

EDIT 2: Wait, you said "how it sold at the double top", but it didn't, at least not at the ~1200 USD double top. It sold on the bounce back, at around 800 USD.

Assuming the indicator is working properly, it shouldn't give a buy signal in 2 weeks.
Although in 2 weeks, in the bullish scenario, we may have broken the 420$ resistance, the indicator
should only give a buy signal in about 2 months, after the corrections from ~520$ to ~400$.
Assuming 275$ was THE bottom, the market moves about 5 times slower than in July 2013.
I do hope that the trading whales will use any information that would speed up the market though.

Have you confirmed your indicator is not repainting? Always looks good in hindsight.. I dont ser how it knew to sell perfectly on the double top as there was no indication that a selloff was coming seems repainting to me...

One algo I had I fell into it by accident analyzing the
jpy daily interventions by boj.. I came across developing an indicator that would show spikes of 100 pips or more within 15 min timeframes averaged across major jpy crosses.. turns out those were capitulation events on larger ti
eframes so I created a bot tobacktest them and found it to be wildly successfull running back to 2007.. howver
fwd testing I came across the first spike but jpy crossed continued on the same path for months longer breaking my rules.. thus leading me to my point that the market never repeats but rhymes. I suggest atleast 2 to 3 years fwd testing before putting any real money in.. or
best to just hold.

Are you ever planning on sharing your secret formula you've discovered?

Hey, I said that much in my OP's Q&A part A good trader will almost certainly outperform this method. It's a (hopefully smart) combination of momentum signals, while a good trader will have seen the 275 drop as an excellent entry point for a mean reversion trading pair, which is probably a lot more profitable.

My algorithm can't do that. It looks at the big swings, and half-way through a  bear market basically says "fuck it, I had enough" That's what I was aiming for, and that's what the algorithm seems to signal quite well, so far. Not that many data points because little trades, so take it with a grain of salt.

From the top of the April 2013 bubble: about 50%. From the 2011 bubble: about 60% even. Not that great if dollar preservation is your highest ranked goal, but like I wrote above, there are 3 ranked goals I had in mind, and riding out a rally for the most part before selling is at the top of that list.

I would probably only set a stop for a buy signal that looks a bit dubious to me (like the one I expect now ). The buy signal come in a bit easier than the sell signals, so it makes sense perhaps to reign them in a bit. How tight depends, again, on your preference for USD preservation vs BTC accumulation, but I'd say at least 5%, up to 10% - tighter than that, and it probably won't work well with the comparably large moves this method seems to accept as "noise" between the tradeable swings.

I understand where you're coming from, but "the bottom is in" and "lazywhale signals buy soon" are two not necessarily related statements.

Like I said in the OP, unless we see a major, continuous drop in price, the method will most likely signal 'buy' in about the next week. I can say that because I can look at the internal variables, and see how far they are towards a buy signal. I can also approximate how far price would need to drop before there will be no buy signal in the short term - that's why I said what I said: barring a drastic price drop, it will signal buy soon.
(EDIT: to be more precise, price would need to fall to a level of 330/340, and stay there for a while, before a buy signal becomes unlikely)

So, 'bottom is in' and 'buy signal coming in' are not necessarily linked because:

a) there is a chance (even if small), that the bottom isn't in, the method will signal buy, and it will be a *profitable* signal nonetheless. That would happen if there is a second bottom after the buy signal, that price immediately recovers from before the method signals sell. Like I said in my OP, this method isn't built to catch the absolute bottom, it aims to capture large swings, so in that case, the signal would work just as intended.

b) the more likely scenario if the bottom *isn't* in: the method will yield an unprofitable signal.

I hope it won't do that, but that's why I can say 'method is close to a buy signal' without committing to anything about the market. I just describe ... okay, admittedly: tease a bit, about what the algorithm is about to signal Maybe that's what you dislike? I was trying to gather a bit of interest by doing that, but it's a bit sleazy perhaps to do so.

Good post as usual oda.

Beginner traders should not be trading frequently. Taking a longer-term view and developing a strategy around it (i.e. this one) is far more effective than day trading, or worse yet, day trading on margin.

I used to trade every other day. Couple months ago I switched to trading every once in a while. Works better for me.
Day trading is fun with a small stack but once your stack gets bigger it is hard to move your coins quick at one price.

My last trades by the way are:
Sell at 480$, Buy at 300$.

I'm all-in again since that huge sell wall on Stamp at $300,- I took a good chunk out of.

There are chances that it won't scale down. While charts are fractal, bitcoin long-term has some special features that IMO are not present at shorter time-frames: exponential growth, orders-of-magnitude runs, 8-month cycles. If the method relies on them, it won't scale.

This algorithm's buy/sell signals look fairly similar to those on the daily ichimoku cloud.

Oda,

If your signal indicator is going to have a buy in a week or two, its pretty obvious that you (or the indicator) believes the bottom is in, that we will retrace off local highs, consolidate, and then start making a move up.

Its not rocket science to eyeball your chart.

Or are you leaving the possibility that the indicator will not offer a buy signal and we will go to new lower lows.