I'm curious how everyone here tests their arbitrage bots before they go live. I'm not talking about backtesting but rather running the strategy using real market data without actually placing any orders.
From what I've noticed, the problem usually isn't with the calculations but more about everything else involved. A setup might look good when you first see it, but then things start to change. Timing gets off, prices fluctuate, and the final outcome can end up being way different from what you thought. Plus, you have to consider how many opportunities you actually take, how the system reacts when part of your balance is already tied up, and what happens if a bunch of similar signals hit at once.
That's often where things start acting very differently than you planned. So, do you test these scenarios in advance, or do you just go live with a tiny amount and tweak things as you go?
This is nonnegotiable, bro!
We are in an advanced technological age where things might not be the way they look/seem. Do your forward-test paper trading simulation with it, and never backtest. This reveals hidden latency and non-existent liquidity. Just ensure your settings are right.
Do this for weeks before proceeding to live testing it. But with a tiny amount first.
You should not use real money to test your arbitrage bot. Use binance spot demo mode to test your bot against real data.
Read up their docs, although they will warn you their demo mode realistic market data is not equal to real data, it's good enough.
After that you can try with small sum to see whether your bot works as intended.
Using real money just to test arbitrage bot is wasting money, you don't even know how it will react to real data beforehand.
Arbitrage bots have not become that profitable, as the price differences between platforms, deposit and withdrawal fees, in addition to the costs of managing the bot, will make the profits almost non-existent. Therefore, even testing them is only done as part of a trading strategy and not separately.
I would concur without any hesitation as to why it's a good practice to always use simulation first before entering the market later. If one intends to trade arbitrage they must remember how high the risk is and as a new entrant testing different strategies of trading besides holding or accumulation, simulation would help to test if your logic is correct.
Simulation helps to achieve an actual average of the EV of your strategy over more trades than you can count and that's why it is important to save yourself from loss.
The question is, is arbitrage still possible if it only works with top exchanges where there are only small price differences? Another issue lies in the platforms that support arbitrage bots. I mean, profitable price differences might be achieved if they work on low-volume exchanges. If simulations without real orders are required on such exchanges, the results will be different after live implementation.
I haven't had any experience with arbitrage bots, but if I were to do so, I would have to test them thoroughly.
I'd be surprised if paid bots don't have testing feature. If we're talking about open sources or free ones, I guess the user will have to setup the data themselves, which is another hurdle for someone looking for a simple tool to use. What about you OP? Is there a tool that you know can run simulation perfectly with little to no manual coding? I don't know how you can simulate a delay on transfer due to network issues, exchanges being slow to update your balances, and so on though. I guess that's a variable that can be removed if your goal is to know how multiple algorithm behaves when the timing is tight.
That's the thing with arbitrage bot, there's not much opportunity even within the exchange itself. Market makers are already there to get all profit from arbitraging between the pairs.
We'd be giving free money to the API provider or the bot maker instead of actually making money. Even within decentralized exchange it's all the same.
Yeah, that is actually a tricky one. From what Ive seen, there is no tool that really simulates all of this properly without making a lot of assumptions, especially when it comes to how execution actually plays out or how exchanges behave under load.
Most tools either simplify those parts quite a bit or just ignore them, so even if they are easy to use, they dont really match what happens in real trading. There are more advanced setups that try to get closer to reality, but they usually take more effort to set up and still dont get everything right. So it usually comes down to a tradeoff. Either it is simple, or it is more realistic, but rarely both.
What's there to test? Arbitrage opportunities either exist or they don't. Essentially, you're buying the spread and closing trades when it goes to zero. A widening spread can wipe out an arbitrageur's deposit. So it's important to know roughly what the maximum spread has been, but you can look that up on historical data without any testing at all.
Most likely, testing is needed to check whether the bot actually follows the algorithm you designed whether it opens and closes trades correctly, and how it behaves when the internet connection drops.
But not all exchanges have a demo trading mode. Mine on Cryptomus doesn't. So I run bots with the minimum deposit on the account. And that's the best way to verify the algorithm is working correctly. On a demo account, there's no slippage, no sudden jumps in bid/ask prices, no price gaps on news, and a lot of other things missing.
If the question is for only yes and no then the answer is yes. You even said it works even for the top exchanges only . But I still think that arbitrage is for all types of exchanges. Simulations are just test-drive kind of thing anyway. So they might still be different or we might be able to adjust things here thoroughly depending on our risk appetite until we achieve that true to life experience that we can think of.
If not, we can also try the real thing but just only use small amounts if possible. Yes mate, everything needs to be tested out first before we use them. That is of course to minimize errors and build more confidence in the process.
Only use it for certain moments, because there is a friend of mine who uses this kind of Arbitrage Bot which has been integrated with all exchanges and
calculates all withdrawal and deposit fees so that it will be found how much profit will be obtained.
Now with the help of AI maybe arbitrage bots can be even more powerful, but of course it won't always provide profits,
because of the difference in withdrawal, deposit, trading fees for each exchange platform.
Obviously if you have any kind of trading bot, whether arbitrage, spot or derivatives then you will need to test it first. Nobody is going to risk throwing their money away on an untested bot. But I do not think that they have to write their own testing system when there are so many testnets available for that very reason.
I even wrote my own BTC/USD spot trading bot for friendly competitions against other traders.
I assume by testnet you mean testing platform for bot trading activity. Is there any free one out there? I haven't heard a platform dedicated for such thing. At best it's an open platform that offers API data so you can backtest or something similar. Doing live trading and making "ghost orders" is new to me. I guess they won't give everything for free. Still interesting to see if the data is anything useful though.
Yeah of course.
For trying out options trading you can try the Deribit testnet for example. There are others as well. However I do not recommend trading derivatives no matter how much you test because it will always end up with losing all your money.
I have not tried others yet but there also testnets for DeFi and spot trading.
Sorry for the late reply, I missed the later posts and thought the thread had gone quiet.
I agree that no simulation can fully reproduce live trading. Demo or test environments can be useful, but they usually simplify execution a lot. Slippage, partial fills, balance update delays, and exchange behaviour can still make the live result very different.
At the same time, going live with a minimum deposit is not perfect either, because you are already testing with real funds.
For me, the point of simulation is not to predict the exact live result. It is more about seeing how the setup behaves before real execution: which exchanges and assets may be worth focusing on, how risk settings, fees, and exchange discounts affect the result, and how balances move after a sequence of trades. That is hard to understand from a few live trades, because the question is not only whether one opportunity works, but how the setup behaves over many signals.
The testnet is a little different; you can see spikes that you wouldn't see in live trading. So, if you use or test an arbitrage bot, you may get a bad result or a good result, but when you switch to live trading, the outcome is negative.
So, I believe the best approach here is to test a small amount on sub-accounts rather than the main account. At least, you can collect some data that you can then use to improve the bot's profitability.
Tbh, most simulation are being used to test whether bot is working as intended or not and find out bugs, also testing against real live data while spending nothing occasionally.
But there is nothing wrong in wasting even $50 for testing your bot if you are sure you can make significantly more money when the bot went live and start printing you money, sort of an investment.
The only thing I test is use small money to test the water. If I am sure with the situation and can handle the risk, I will entering the market and vice versa. The demo will be different on the real so you need to set up carefully because the market can change after your setup finish. This makes you create more strategies anticipating if the situation change so you can adapt it.
How does that works exactly? You run your bot on a platform like Binance & other exchanges, use $50 to run your algorithm and see if it produces the profit you wanted after making several orders or so? How much money is required for transfer then? At what point do you think your bot is running fine and you can use it freely? Sounds like you'll need a decent amount of money if arbitrage with bots is the goal.