Whether a late-executed signal is worth chasing cannot be decided by gut feeling. You need to run a delayed backtest first—put the "lateness" condition into your simulation and see what remains of the signal after subtracting the real cost of delay. Academic research already gives a sobering benchmark: in on-chain trading scenarios, adding just 1 second of delay turns a barely positive return directly negative. But that does not mean all signals should be abandoned. It means latency itself is a cost, and it must be measured in the same framework as fees and slippage. If a delayed backtest shows that returns go to zero or turn negative after subtracting delay, there is no basis for chasing the trade. If positive returns remain and you can confirm that this delay is reproducible in live trading, only then is it worth executing.
A leading global cryptocurrency platform,suitable for both beginners and experienced traders.
New user benefit: 20% off trading fees upon registration!！
Delayed backtesting answers: "At what price does the signal actually get filled?"
A normal backtest implicitly assumes that the moment a signal appears, you can get filled at that exact price. In live trading, that is impossible. Between seeing the signal and placing the order, exchange matching, and order confirmation, there is network latency, manual operation time, and platform processing time. During this window, the price may have already moved against you.
The delayed backtesting approach is straightforward: separate the signal timestamp from the actual fill timestamp. The signal triggers at time T, but the fill price is taken from the first available price after T plus the delay duration. Backtesting tools usually implement this with a shift parameter. For example, if you set a 2-minute execution lag, positions will be settled at the price 2 minutes after the signal.
The key point is that the delay value cannot be a random guess like "a few seconds." It should come from observing your actual execution path. If you manually follow signals, the time from seeing the notification to completing the buy click could be anywhere from tens of seconds to several minutes. If you use automated API execution, the delay may be under one second, but network jitter and exchange rate limits will still cause variation. The delay value you enter in the backtest should be close to the most common delay you actually experience, not the fastest or slowest extreme.
Academic data: 1 second of delay is enough to flip the result
An empirical study on on-chain trading signals provides a useful benchmark. The researchers built 369,036 decision points, trained a model to identify optimal entry timing, and then tested it out-of-sample. Under ideal conditions with no delay, no fees, and no slippage, the highest-ranked decisions averaged a return of about +0.054 (in a certain return unit). After adding observed protocol fees, the return remained essentially unchanged. But after adding 1 second of delay, the average return became -0.099, with a 95% confidence interval of -0.114 to -0.083 and a median of -0.083. The win rate turned negative. When the delay was increased to 3 seconds and 5 seconds, results did not get significantly worse, but the negative return state persisted.
What this data shows is that for signals relying on short-term price movements, delay destroys returns not by "earning a little less," but by directional reversal. The brief advantage window captured by the signal can close within 1 second. In out-of-sample testing, the highest-ranked decisions under the project cost model had returns between -9% and -4%, and no group's confidence interval upper bound exceeded zero.
Another study on high-frequency trading on Bybit and Binance provides complementary evidence from the execution side. The research found that delay leads to adverse selection: orders that later turn out to be profitable have a higher probability of failing to fill or filling at worse prices due to order book updates, while orders that later turn out to be losing almost always fill at the expected price. This means delay does not affect your fills randomly—it systematically filters out good trades and leaves you with bad ones.
How to run your own delayed backtest
You do not need to write your own backtesting framework. If you use an existing backtesting library, find the shift or execution_lag parameter and separate the signal timestamp from the fill timestamp. If you do not have a ready-made tool, you can use a simplified process:
Step 1: Determine your actual delay distribution. Use a stopwatch to record your signal execution time: from the moment the signal appears on your screen to the moment the order is confirmed as filled. Record at least 20 times and take the median. If you use automated execution, check the timestamp difference between when the strategy sends the signal and when the exchange returns the fill confirmation.
Step 2: Use this median as the fill delay in your backtest. If your backtesting tool supports variable delay, that is even better, but a fixed delay will already expose most of the problem. The key is to compare two sets of results: zero delay (ideal fills) and your actual delay. The difference between the two is the cost that delay imposes on your strategy.
Step 3: Look at delay cost together with fees and slippage. If you earn 10% with zero delay and 3% with your actual delay, that 7% gap is not "acceptable friction"—it means your live return ceiling has been compressed by 70%. If you were only earning 3% with zero delay, the delay might push you to zero or negative, and chasing the trade would have no economic meaning.
Step 4: Check whether delay changes the signal's win rate and profit/loss ratio structure. Some signals have a 55% win rate and 1.5 profit/loss ratio with zero delay, but after delay the win rate drops to 45% while the ratio stays the same. This kind of change tells you more than the total return number about whether the signal has lost its edge source under delay. If the signal's edge comes from "getting in before others," delay will kill it directly. If the edge comes from "holding long enough for value to revert," a delay of seconds to minutes may have very little impact.
When chasing a signal may still make sense
Delayed backtest results are not a blanket "abandon" or "chase" verdict. There are several situations worth further judgment.
The signal's time horizon is long. If the signal points to price movements over hours or days, and your delay is tens of seconds to minutes, the delay accounts for a very small proportion of the holding period. In the academic study mentioned earlier, the on-chain signal holding time was only 20 to 60 seconds, so a 1-second delay accounted for 1.7% to 5% of the holding time, magnifying the impact. But if the holding period is several hours, a 1-minute delay only accounts for 0.3%, and the impact will be much smaller. When backtesting, you should evaluate the relative impact of delay based on your actual holding period.
The signal itself already includes the delay. If the signal publisher you follow has already executed the trade before publishing, there may already be a gap of minutes to hours between the "signal time" you see and the "actual entry time." In this case, you need to judge whether the price has already deviated significantly from the level at publication time by the time you see the signal. A Bitget case description shows that the same signal reacted to in real time versus after a 3-hour delay could differ by +11% versus -24%. This case comes from trading community content, not academic research, but the mechanism it points out is worth noting: the impact of delay depends on how far the price has already moved after the signal was published.
You can confirm that the delay in your backtest is reproducible in live trading. If you put a 1-second delay in the backtest, but in live trading network issues often turn it into 10 seconds, the backtest result is overly optimistic. The delay value in your backtest should be conservative—take the slower end of your actual delay distribution, not the median.
If the backtest shows no profit after delay
Not chasing the trade is a reasonable conclusion. The value of delayed backtesting is not to find a parameter that "still makes money," but to let you know before you commit capital: does this signal still have positive expectancy under your execution conditions after subtracting delay costs? If the answer is no, chasing the trade only provides liquidity to the signal publisher or to faster traders.
If the delayed backtest still shows a weak positive return, but the confidence interval crosses zero, it is also not worth chasing. A return too weak to distinguish from zero will be easily wiped out in live trading by factors you cannot model—exchange outages, network fluctuations, or the few seconds you hesitate.
A leading global cryptocurrency platform,suitable for both beginners and experienced traders.
New user benefit: 20% off trading fees upon registration!！
References
- arxiv·No better than random entry, 2026-11-02
- ebt-backtest·PyPI, 2026-10-28
- Bitget Insights·REAL-TIME VS DELAYED SIGNALS: WHY TIMING CHANGES OUTCOMES, 2025-10-17
- Queen Mary Research Online·The Good, the Bad, and Latency: Exploratory Trading on Bybit and Binance, 2026-11-02
- Oxford University Research Archive·Latency and adverse selection in crypto taker orders, 2026-11-02
