AisTradingDo AI trading signals work? How to validate them in simulated mode before putting in real money
Team at AisTradingWritten with AI assistance20 min readIn summary
- An AI trading signal is not a prediction: it is a proposal with entry, stop, target, and risk/reward ratio written before entering. If the stop is missing, it is not a signal: it is an opinion.
- The confidence score (in AisTrading, from 60 to 95) is used to rank candidates within the same analysis cycle, not to promise accuracy. A 90 is not "90% probability".
- The only honest way to know if a signal system works for you is to see it operate in simulation for weeks, with your capital and your level of risk, without touching the settings midway.
- Two good weeks mean nothing. Look at number of trades, expectation in R, maximum drawdown, and worst losing streak before looking at the loose win rate.
- Risk management weighs more than accuracy: a mediocre strategy with limits per trade, daily loss cap, and firewalls survives; a good one without caps does not.
If you have arrived here asking "Do AI trading signals work?", it is likely that you have already seen screenshots with curves that only go up and impossible accuracy percentages. Distrusting is the correct reaction. This article does not try to convince you that signals are accurate: it explains the method to verify it yourself, and how to do it with simulated money in AisTrading, the signal and risk management app from AIsToHire, before putting in a real euro. AisTrading does not manage your money or guarantee results; trading involves the risk of losing part or all of your capital.
Everything described below are mechanisms that the app has implemented and default values of its engine as of October 2026. There are no profitability figures in this article, neither actual nor simulated: it is a deliberate decision, because publishing simulated results is financial promotion and is exactly what the services you should avoid do.
What is a trading signal really? Entry, stop, target, and nothing more
A signal is a trading proposal with four numbers written before entering: entry price, stop price (where you admit you were wrong), target price, and the relationship between what you risk and what you seek. Everything else is narrative.
In AisTrading, each signal is stored and displayed with those fields, plus the context that generated it:
- Entry, stop, and target in price, not in vague percentage.
- Risk/reward ratio (R:R) calculated from those three levels.
- Confidence score between 60 and 95.
- Reasons, with the specific value of each condition that has been met (breakout of the opening range, price above the VWAP, volume against its average, RSI, position relative to the 200-session averageā¦).
- Detected risks for that same trade.
- Proposed size in units and amount, and the risk in euros that it entails if the stop is triggered.
- Expiration: a signal does not last forever. By default, it expires in 2 hours; those waiting for a limited entry below the close have a longer window (32 hours) and, if the price does not reach the limit, they expire without executing.
The practical consequence: if a service sends you "BUY BTC š" without a stop, without a target, and without telling you what condition has been met, you cannot validate it even in theory. There is nothing to measure.
What does a confidence score of 90 mean? It is a ranking, not a promise
The confidence score ranks the opportunities of the same analysis cycle from best to worst according to the conditions each one has met. It is not a probability of success, and treating it as such is the most common mistake of the signal reader.
This is how it works in the AisTrading engine, and why the number has that role:
- In each cycle, the engine evaluates the entire universe of enabled symbols and accumulates the candidates that meet the rules of some active strategy.
- Then it ranks them by confidence; in case of a tie, it decides the best risk/reward ratio and, if the tie persists, the relative volume.
- It distributes the available slots of the day from top to bottom (by default, 3 trades), deducting those that do not enter due to exposure limits.
- Each signal keeps its position in the ranking and how many candidates were competing in that cycle, and the distribution remains in the activity log.
In other words: the score answers "of what is available today, what is the best?". It does not answer "is this going to win?". That is why the ceiling is 95 and not 100: the maximum is a ranking, not a certainty. If you ever see a service that gives you "97% probability of success", you have just found your first warning sign.
How does a 24/7 signal engine analyze the market and why is it not an oracle?
The AisTrading engine is a process that wakes up every 5 minutes and goes through a fixed cycle: evaluating alerts, managing open positions, expiring old signals, checking pending limit entries, and scanning each enabled market. It is a clock with rules, not an intelligence that "intuits" the market.
There is a distinction that almost no one explains and that is important to understand:
- How often the engine wakes up (every 5 minutes) is not the same as what candles it analyzes. The analysis time frame is configurable and by default is 1 hour for stocks and crypto; some strategies use daily candles because that is the horizon in which they make sense.
- That separation exists for a measurable reason: with 5-minute candles, the stops are so tight that they end up below the cost of the operation. In crypto, Kraken's taker fee (0.26% per side, or ~0.52% round trip) plus spread and slippage eats up a 0.25% stop before it even starts. An operation that begins losing is not a signal: it is a fee.
Each active strategy is a hypothesis about how the market behaves, and it fails when that hypothesis is not fulfilled. The ones provided by the engine by default, stated out loud:
- Panic rebound (RSI 2): assumes that an extreme and short-term drop within an upward trend tends to bounce back. It implements a rule published by Connors and Alvarez (2008), with an entry limited below the close and exit by indicator. It fails in drops that do not bounce: sustained downward trends.
- Weak close (IBS): assumes that a close near the day's low recovers within the range. It fails when the weak close is the beginning of something worse.
- RSI reversal and opening range breakout (ORB): mean reversion and continuation of momentum. They are opposing hypotheses; when the market rewards one, it punishes the other.
- End of month and hangover Tuesday: calendar patterns, limited to the indices where they have been measured. They provide opportunities spread throughout the year, not magnitude.
- Intraday satellite portfolio (Luna), with assigned own capital and in simulation while the internal meter decides whether to stay.
This is what it means "multiple strategies analyze the market": multiple hypotheses competing each cycle, each with its blind spot. An honest system tells you what that blind spot is. A snake oil salesman tells you it works "in all markets".
The TEST mode is not a toy demo: it is your testing ground
The TEST mode of AisTrading exists to audit the system, not to entertain you. Each market has its own mode, and in TEST you operate with simulated money on real prices: crypto against the simulator fed with Kraken prices, and stocks against Alpaca Paper, where the engine sends real bracket orders to a free test account of the broker.
Details that make the difference between a serious simulator and a decorative demo:
- Costs are modeled and saved. Each simulated position records commission, spread, and slippage, and the same cost model is used in backtests. Without costs, any system seems profitable.
- The source of the data is labeled. The app marks where each price comes from and alerts when the public feed is delayed (up to about 15 minutes). What does not have a contracted provider is declared as unavailable with its reason, instead of being filled with invented figures.
- They do not give you a false history. At the end of onboarding, the app does not plant a track record of trades that never occurred: it starts with the portfolio at its initial capital and the indicators at zero. Sample data exists, but only if you request it manually in Settings ā Test Data, and they are labeled as simulated.
- The backtester and the engine share the same evaluation code, including the trimming for exposure. A backtest that the app itself could not reproduce while operating is worth nothing.
And the detail that exposes any signal service: switching to real money in AisTrading is not a switch. You need to have the real broker connected, the credential test to pass, and write the word REAL on a screen that first shows you your own TEST statistics (trades, accuracy, profit factor, and maximum drawdown) and warns you if your history is short. Going back to TEST does not close any real positions you may have open, and the app tells you so.
How many trades do you need before concluding anything? Many more than you think
Two good weeks are not evidence of anything. With few trades, in a single market regime and with the bias of having chosen the moment to look, luck explains the outcome better than strategy. The app shows you on screen: below 10 closed trades it warns of small sample, and before letting you activate real mode it flags the history as short if you have fewer than 20 trades or an accuracy below 50%.
These warnings are minimum thresholds, not the goal. For its own automatic cut-off mechanism to dare to deactivate a strategy, the engine requires 100 measurements of that strategy. That is the real order of magnitude for validation.
A validation protocol that does work, in seven steps:
- Set the configuration and donāt touch it. Risk per trade, limits, active strategies, and time frame. Changing parameters midway invalidates the sample: you would be measuring three different systems and averaging them.
- Put in the capital you are really going to use. Validating a system with ā¬100,000 simulated that you are going to operate with ā¬2,000 does not measure the same: the minimum size, the cut from exposure, and the weight of costs change.
- Define in advance the minimum number of trades you will require before deciding, and write it down. If you decide the criterion after seeing the results, the criterion is the result.
- Require at least one adverse period. A month of continuous rise does not prove a mean reversion system. If you havenāt seen how it behaves in a correction, you havenāt seen it.
- Measure by strategy, not in bulk. A positive aggregate may be hiding a strategy that constantly loses and another that covers it.
- Note the trades that were not executed and why (daily limit, exposure, kill switch). It is part of the system: a strategy that generates 40 signals a month of which you can only take 3 a day is not the same strategy.
- Decide with written rules before starting. "If after N trades the expectation is not positive, this strategy is out." Without that written phrase, confirmation bias decides for you.
What metrics to look at and which to ignore
Look at the number of trades first, the expectation in R next, and the maximum drawdown and the worst losing streak before celebrating anything. The standalone win rate is the favorite metric of those who sell signals because itās the easiest to manipulate: just close the winners very early and let the losers run.
The ones that matter
- Number of closed trades. It is the sample size. Without this, the rest are anecdotes.
- Expectation in R (average result per trade measured in units of risk). It is the honest metric: it answers "How much do I win or lose on average every time I risk 1?" AisTrading calculates it per trade and per strategy.
- Profit factor (gross profits divided by gross losses). Useful, but very unstable with few trades.
- Maximum drawdown on the capital curve. It is the real question: how much do you have to endure at the worst moment without shutting it all down? An exhibited return without its drawdown is half the information, and the missing half is what takes you out of the market.
- Worst consecutive losing streak. No one abandons a system for its annual profitability: they abandon it on the seventh consecutive loss. Itās good to know in advance that this streak exists.
- Total accumulated costs. In short-term trading, they are often the difference between winning and losing.
- Average risk/reward ratio and average duration of trades: they indicate whether the system you are operating is the one you thought you were operating.
AisTrading displays these metrics on its history and track record screen, allows filtering by strategy and by market, exports trades to CSV for analysis outside, and runs the results through a Monte Carlo simulation to see what range of streaks is compatible with your sample.
The ones that shouldnāt move you
- The isolated win rate. A 70% accuracy with losses three times greater than the gains is a system that loses money.
- The best trade. It is not repeatable and is often the sales argument.
- A snapshot of the capital curve without the number of trades, without period, and without drawdown.
- Accumulated profitability without the risk needed to achieve it.
Why risk management matters more than being right
A mediocre strategy with limited risk survives a bad streak; a good strategy without limits does not survive its first bad week. The correct order of concerns is: how much do I lose if I am wrong, how many times can I be wrong in a row, and only then how much do I get right.
The limits that AisTrading brings configured by default (engine values, October 2026):
- Risk per trade: 1 % of the reference capital. Above 2 %, the app explicitly warns that it accelerates losses in a bad streak.
- Daily loss limit: 2 %. Once reached, the engine stops opening positions that day.
- Weekly firewall: 5 % and maximum drawdown firewall: 10 % (configurable between 5 % and 20 %).
- Daily trade cap: 3. It is the limit that turns "the system has given 40 signals" into realistic trading.
- Maximum exposure: 15 % per symbol and 30 % per sector. The size of each position is trimmed to the free exposure gap before opening, not rejected afterward.
- Kill switch: a switch that immediately stops new entries in both markets, with the option to close everything open. Alerts continue to function with the kill switch on.
- Relaxing a protection requires confirming the risk and is logged in the activity log, with a button to restore the recommended values at once.
None of this makes a system make money. It does something different and more important: it guarantees that a bad period is a bad period and not the end.
Forward testing: how an honest system measures its own edge and shuts itself off
The difference between a backtest and a forward test is that the backtest is done on the past that you already knew when you wrote the rules. The forward test measures the signals that the engine has emitted after, on a market that no one had seen.
AisTrading is set up this way:
- For each strategy, it calculates the expectation in R by combining the actual result of the positions it opened and the theoretical result of the signals it generated and were not executed (resolved in the time frame specific to each strategy and with its exit rule).
- Signals with simulated data or from manual sources are excluded from the calculation.
- When a strategy accumulates 100 measurements or more and its expectation is zero or negative, the engine deactivates it automatically, notifies it, and logs it in Activity. It cuts once; if you decide to reactivate it, it respects your decision.
- The track record screen shows the status of each strategy with a clear label: measuring, with an edge, without an edge, or deactivated.
This is, in our opinion, the only honest "guarantee" that a signal system can give: not promising accuracy, but measuring itself live with a pre-set criterion and shutting off what does not demonstrate an edge.
Six warning signs of a signal service that is not serious
If a service meets any of these six, you don't need to analyze its signals: you already know what it is.
- Displays returns without the maximum drawdown that was needed to achieve them, nor the period, nor the number of trades.
- No verifiable history. Screenshots, messages from a deleted channel, or a "history" that starts just where it is convenient. If you cannot export the trades and review them one by one, it does not exist.
- Does not have a simulated mode, or it has a demo cut that does not reflect costs or real execution.
- Pushes you to real money from day one, with urgency ("limited spots", "the market does not wait"). A system that trusts its method is not in a hurry: it lets you validate it for weeks.
- Does not model commissions, spreads, or slippage. In short-term trading this is not a mere accounting detail: it decides the sign of the result.
- Promises to manage your money or guarantees results. That is not a signal service; and if it is not authorized to manage third-party capital, it is also a regulatory problem.
Add a seventh, more subtle one: services that present a success rate as their main argument. It is the metric that withstands analysis the least and the one that is best polished.
What AisTrading does not do, spelled out clearly
AisTrading is a tool for analysis, signals, and risk control. It is not a manager, it is not an advisor, and it is not a money-making machine.
- Does not manage your money nor makes decisions for you with real money: in REAL mode, automatic execution is disabled by design. The engine proposes, you execute.
- Does not guarantee results nor issues market forecasts. Nothing that the app shows is financial advice, and trading involves the risk of loss, including total capital loss.
- Does not publish its own or simulated returns, by express decision: the numbers that matter are the ones you generate in your own TEST mode, with your capital and your configuration.
- In crypto, it does not operate alone by default: it generates and proposes signals, and you decide the execution. This can be changed in Settings, with knowledge of what it implies.
- Does not invent data it does not have. What it does not have a contracted provider for (economic calendars, social sentiment, institutional flow, among others) is declared as not available, with its reason.
Frequently Asked Questions
Do AI trading signals work?
It completely depends on the specific strategy, the cost of trading, and the risk management that accompanies it, and no one can honestly answer that for you. What you can do is check it: let the system operate in simulated mode with your capital and your risk level for weeks, measure expectation in R, maximum drawdown, and worst streak, and decide with a written criterion beforehand. Anyone who answers "yes, an 85% success rate" is selling you something.
How long should I test signals in simulated mode before switching to real money?
Time matters less than the number of trades and the variety of market conditions. AisTrading warns of a small sample below 10 closed trades and marks the history as short below 20 before allowing you to activate real mode; its own automatic cutoff mechanism requires 100 measurements per strategy. As a practical rule: several dozen trades per strategy and at least one adverse period within the sample.
What is the confidence score of a signal and can I trust it?
It is a value between 60 and 95 that ranks the detected opportunities in the same cycle according to the conditions met by each one. It serves to prioritize when there are more candidates than available spots on the day and to filter alerts. It is not a probability of success and should not be read as such.
Do I need a broker account to validate signals without risking money?
Not to start. In AisTrading, crypto in TEST mode works on a simulator fed with real prices from Kraken without the need for credentials. To validate actions with realistic execution, it is advisable to connect a Alpaca Paper account (free and with simulated money): the engine then sends real bracket orders against that test account. Without credentials, stock trades are simulated and labeled as such.
What metrics should I look at and which should I ignore?
Number of trades, expectation in R, profit factor, maximum drawdown, worst streak of consecutive losses, and accumulated costs. The one to avoid looking at in isolation is the success percentage: it is compatible with a system that loses money and is the easiest to disguise by closing winning trades early.
Why doesnāt an honest service publish its profitability?
Because profitability without its maximum drawdown, its period, its number of trades, and its cost model does not inform: it persuades. And because publishing simulated results as a commercial argument is financial promotion. In AisTrading, the decision is deliberate: the method for measurement is visible, and the numbers are generated by you in your own simulated mode.
Does AisTrading trade for me if I connect my real broker?
No. In REAL mode, automatic execution is disabled by design: the engine analyzes, proposes the signal with its levels, and it is the user who decides to execute it. Automatic execution only operates in simulated mode. And switching to real mode requires a connected broker, correct credential verification, and writing the word REAL on a screen that previously shows your test statistics.
What happens if a strategy stops working?
It is the norm, not the exception: each strategy is a market assumption and stops working when that assumption is not met. That is why the engine measures the expectation of each strategy with its own live signals and, when it accumulates enough sample without an advantage, it deactivates it itself and notifies you. You should demand the same from any service: a written shutdown criterion before starting.
How to validate this in AisTrading, without real money
If you have made it this far, you already have the method. AisTrading is the app from AIsToHire that allows you to execute it: signals with entry, stop, target, risk/reward ratio, and the numerical reasons for each; an engine that analyzes the markets in 5-minute cycles with several competing strategies; simulated mode by market with real prices and modeled costs; configurable risk limits and firewalls; and live measurement that deactivates what does not demonstrate an advantage. It is at AisTrading ā AI Trading Signals Explained, you start in test mode without connecting any broker, and what you decide to do next with real money is your decision, with your risk and under your responsibility.
What you are not going to find in AisTrading is a profitability figure that convinces you. You will find the tools to produce your own figure, with your capital and your configuration, and the method to know if it means anything.
Validate the signals in simulated mode, without real money
Validate the signals in simulated mode, without real moneyKeep reading
View all articles ā
Agent-based trading: what risks you take when you let an AI agent operate with your money

Why Your Car Insurance Increased in 2026: Average Premium of ā¬1,412 and Historic Record

Discussing via WhatsApp or talking in person: what science says about couple conflicts through messages
