Wouldn't it be nice... (the LazyWhale algorithm)

[windjc]

mid-term trades are the best

Oh yes; although I use like 20% of my stash for that still gives enough profit to have a coffee.. or two.. or a car. But you do need lots of patience. Some kind of robot to tell you what to do may help discipline, so GJ oda, interesting what your robowhale has to say (although if I were it I'd be screaming buy buy since last week).

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.

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

[1714178696]

1714178696

[1714178696]

1714178696

[1714178696]

1714178696

[1714178696]

1714178696

[1714178696]

1714178696

[sickpig]

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

Q: Are you going to tell us the exact parameters of the method?

A: Not right now. Think of this as a 'public test' of the method, to see how it performs. If it fails, you're free to ridicule me publicly as well Cheesy

[sidhujag]

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.

[Tzupy]

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.

[oda.krell]

Double answer:

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

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.

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.

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.

[sidhujag]

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.

[vipgelsi]

Not a bad strategy.

[oda.krell]

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.

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.

[fewcoins]

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.

[oda.krell]

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.

[sidhujag]

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.

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.

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.

[fewcoins]

This is hilarious... I love how everyone has to keep referring to bitcoin's past performance
GLTA

[Raystonn]

You are right.  Let's discuss its future performance.  Let me get my crystal ball.

[lyth0s]

This is hilarious... I love how everyone has to keep referring to bitcoin's past performance
GLTA

I originally thought this fewcoins account was another falllling shill account, but there are some subtle differences.

[Odalv]

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.

or 0 BTC and 0 USD if traded at mtGox

[Wandererfromthenorth]

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

[oda.krell]

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.

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.

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.

Think I understand it now. Most trends used in the signal are long term enough that this doesn't appear to happen easily, and I'm using a threshold that probably reduces the likelihood for it happening further.

I'll test it on other charts / markets as soon as I have time. Right now, I want to add one small feature, a "built in" mildly volatility adjusted stop loss, and see how it performs over the history with that built in stop compared to the current stop free version.

[NotLambchop]

...
I'll test it on other charts / markets as soon as I have time. Right now, I want to add one small feature, a "built in" mildly volatility adjusted stop loss, and see how it performs over the history with that built in stop compared to the current stop free version.

[oda.krell]

... a "built in" ... stop loss

+

=

[oda.krell]

New version (V3.2). Updated to include (rather "lax") stop loss triggers. I backtested a range of stop percentages to make sure they don't interfere with the signals (and corresponding trades) in the past. The point was not to optimize historical profit further (which would be cheating imo), but adding a level of safety to the output of the method. Exact percentage values at which stops are set changes, but roughly, around 10%, 15%.

Suggested position still USD. Here's an updated view, with the approximate area in which either a buy signal is triggered (green), or a change to a buy signal becomes unlikely in the short to mid term (red).