Any update on this HCP?
Yeah, sorry... been working pretty hard over the Xmas break.
guigui371 was also running the same experiment. It seems he had a similar period of inactivity before the bot woke up again on Christmas Eve and made another
poor trade for another large loss. I assume you have therefore suffered similar losses.
Pretty much... it was a total shit show... as I'm lead to believe, their were FOUR different outcomes which was apparently dependent on what VPS your bot was running on??!?
So, some users ended up with a 1% win... some users got hammered for a 30% loss... and a couple of results inbetween.
I ended up with a ~25% (-0.01943218 BTC) loss... for a total loss now of -BTC 0.03928640 (-39.29%) since my original 0.1 BTC deposit on Nov 28:
The bot was taken offline for maintenance... and hasn't been trading since.
Seems that those on a "good" VPS, ended up with a trade being filled (for a couple of small amounts like "139" and "143")... and a (fairly reasonable) stoploss set around -$300 ($7238) from the market rate at the time (was around $7500-7600)... those on "bad" VPS, didn't... and turned in the small 1% profit. The market then dropped, and sat there for around 12 hours or so... bouncing around the $7300 mark... which in and of itself, wasn't exactly a bad thing... what WAS a bad thing however, was the bot deciding that it needed to go "full retard" and launched into massive 25x leverage trade within $70 of the previous stop loss that was still active!!?!
You can see here how the bot decided that we suddenly needed close to "13,000" worth of contracts at prices within $70-$90 of the existing stoploss at $7238, and how everything got dumped when it inevitably got triggered:
Obviously, in a market as volatile as BTC... trading so close to a stoploss like that is asking for trouble... and inevitably, the stop loss from the earlier trade got triggered... which then liquidated EVERYTHING
Cue massive losses.
If the bot hadn't opened that "new" position... the loss would have only been ~ -0.00100000 BTC... inside we got slaughtered and much much much MUCH "salt" on the Telegram group... and eventually this response/"explanation" from CryptoSparks:
Hello people, looks like we got an ugly Christmas gift but still an interesting one. It took me a while to read all logs of all bots so here is what happened.
After closing 3 exits in the 7.5k-7.6k area with +1% profit, some users suffered total and then partial VPS slowdowns caused by attacks on vps provider’s servers, that made them take a separate road. A missing entry fill allowed some users to close this trade in just a few minutes and go in stasis with full profit. Others got that extra fill but delay in placing exit caused a missing exit and stayed in the trade with extra risk. Remember that Arakne’s trade ends when last contract is dropped, so a missed fill, may that be an entry or an exit, can influence the final outcome.
Anyway price collapses and we get stuck in a low risk (1%-2%) long at 7.5k. But our stop holds the bart until detected support formation and breakout from it at 7.3k which Arakne bought and averaged down our entry from 7.5 to 7.3 so we could have exited on the 7.4k impulse. Problem was that when entries were added, stop was still considered strong so no update was needed. When the stop algorithm detected possible shakeout, we were already filled at max risk capacity and attempts to edit the stop in a safe zone failed.
The shakeout got our stop for 20$ right before starting the 7.4k “scheduled” impulse that triggered our strong adds. OUCH!
YES, Arakne uses some martingale mechanics, but in a controlled way to average the entry and avoid getting trapped in a trade. That’s actually common among marketmakers, and whatever you like it or not you can have FULL CONTROL on the risk. Long term users may have noticed how Arakne’s avg entry is always a key level in the market range, it simply how markets work. That’s why Arakne starts with leverage BELLOW 1 and reaches max 25 out of 100. We rarely go above 8, but being able to reach 25 is like having always one last ace to play when trapped in a trade. This time Arakne failed in playing it well, because indeed needed to add to position but also to increase stop size.
NO, is just not true that Arakne makes only small wins. For example the day before the first loss we had a +15%.
NO, is just not true that Arakne only makes big losses. Here the loss list of current version since 2018:
0 -1.0239223164147369
1 -7.770849221211668
2 -6.0998134976693805
3 -1.9440504813533817
4 -2.915077465267908
5 -4.060302922521566
6 -1.7726049435660876
7 -3.7430129787980335
8 -14.309137251004552
9 -6.345744077310631
10 -0.009555729922808811
11 -1.652253475973144
12 -1.8721211623094127
13 -1.9481837015325076
14 -8.124456834850443
15 -23.46791646786238
THE PROBLEM with last trade is how savagely she went in buying 7.3 without FIRST updating the stop at 7238 and taking that extra life space mandatory after surviving Bitcoin dipping. The shit happened for a combination of factors such as:
- strong long breakout detected with several supports to buy but at relatively small spread from each other .
- small spread from stop
- big spread from entry
- several algos approving the call
result is that risk was prioritised in averaging down the entry FIRST at all costs. How algorithms gets their priority is highly based on settings managed by machine learning. So i understand you guys are mad because she went crazy near the stop, but the function to increase stop size was there, if she chose not to use it means that statistically in the past that move worked. When shakeout alarm triggered it was already too late and got stuck in that stop. So we are talking about a missing logic part. She needs a way to increase stop when maxed out or the flow stops there, in another and more dangerous trap!
Anyway Stop gets triggered by 20$ right before the impulse’s start to 7.4 that Arakne was expecting as our best chance for exiting the trade . And with the trade setup she was doing, the win would have been absolutely massive(above 30%). DOUBLE OUCH!
The point of failure was that the probable shakeout signal triggered right after we hit max fill capacity, at that point she couldn’t put more risk on the table by moving the stop further.
Getting shakedout is something we cannot ever RISK. So this scenario showed that we need an EMERGENCY TRAILING STOP algorithm that manages this risky situation where we needed to add in order to exit the trap but ALSO we urgently need to increase stop size without risking over the max drawdown allowed to avoid the shakeout. That is already being done by Arakne but with some mathematical limitations. Some fixes to the MM algorithm will make sure that in a way or another, we will 100% of times have the possibility to safely get some ‘survival space’ when is needed. There are different ways to achieve that, for example dropping some levels at loss and use that risk for increasing the stop and dynamically keep it at safety distance or “save some risk” for emergencies.
And for the ones that say that i adjust graphs, look how the backtest engine shows the same loss. I bet that if a loss similar would have already happened in the 2 year of data analysed , the machine learning optimisation would have differed and so the trade result. Is about teaching the algorithms but without giving them too much freedom.
WHAT NOW?
Our vision is to make algorithmic trading accessible to everyone, and a couple losses will not prevent us from delivering the final goal, especially since both losses made the products better.
You are free to leave and yell scam after a loss, but we are here to stay, this is just the beginning.
We care about our users because our business model cannot succeed if users don’t succeed. That’s why most competitors ask for high entry fee while we will be a free to try SaaS. To further reward our early adopters, both losses that hit Arakne in december will be transformed in credit on our platform since day 1.
Just register and send us your nickname in private to receive the credit, then connect different Bitmex accounts via API and spread the capital among all bots based on their risk indicator! With that free credit you will be able to safely recover and start grinding profit again.
As for now you are 100% in one bot, as in any investment, differentiation is the key. Even though you have seen Arakne keeping 100% win rate for 3 months, trading is not about always winning but about managing risk properly over a long period of time. Last trade clearly showed a point of weakness in Arakne’s risk control that absolutely needs to be patched, that’s why i immediately put it in maintenance.
Stay tuned for updates on launch date, meanwhile Arakne will get updated and be back running on your accounts in the next days, so no action is required until platform launch.
Lastly I understand that losing this way is frustrating , but have some consistency. ZakZak just a couple weeks ago you sent me a beautiful message full of thanks and stuff like that, now Im not only a scammer but also the bot doesn’t exits ?
Even tough the new Arakne will start on a new public trading journey, i will also let it manage the previous one account that after peaking at +57% is now down at -18% to demonstrate how recovery is not only possible but an inevitable consequence of proper risk management over long periods of time.
NOTE: CryptoSparks has announced that all users will get next months VPS "free" due to the large amount of maintenance this month and no active trading etc.
A lot of users have pulled their funds and cancelled their API keys etc to cut their losses etc... and the bot is currently in "maintenance". But, I promised I was going to see this through (and I pretty much wrote my 0.1BTC off the moment I deposited it anyway
), so my 0.06071360 BTC shall stay there, API keys are still active and we'll see how it all plays out.
Who knows? Maybe I'll claw back my losses before next Christmas!
NOTE: Basically, imho, the bot was/is "broken". Trading that close to the existing stoploss was reckless. I would not recommend that anyone put money into this bot (or "The Platform"™).