Candle interpolation is the practice of filling a gap in market data by creating one or more price bars that were not observed in the underlying market.
Candle interpolation is the practice of filling a gap in market data by creating one or more price bars that were not observed in the underlying market. In a candle series, each bar normally summarizes actual trading activity over a fixed interval. When data is missing because of an outage, packet loss, or incomplete venue coverage, an interpolation routine can insert replacement bars so the time series remains continuous. Those inserted bars are synthetic. They are generated from surrounding observations rather than recorded from actual trades.
In cross venue research, this distinction matters because continuity and truth are different properties. A continuous series is convenient for charting, feature engineering, and model input. But if the continuity is produced by interpolation and not by real prints, the resulting candles describe an assumed path through time rather than an observed one. Sonar Sciences notes in its cross venue data research that market data from different venues must be aligned and normalized carefully because venue specific gaps and timing differences can contaminate comparisons. A synthetic bar created to smooth one venue through a gap can therefore create a false impression of price agreement, lead lag structure, or spread behavior across venues if it is treated as equivalent to an observed bar.
The mechanism is simple. Suppose a one minute series has valid candles at 10:00 and 10:03, but the 10:01 and 10:02 intervals are missing. An interpolation procedure may construct the missing bars by carrying forward the last known price, linearly stepping from the 10:00 close toward the 10:03 open or close, or inferring open, high, low, and close from neighboring candles. In every case, the inserted values are model outputs. They are not the result of trades that actually occurred during those missing intervals. This means the synthetic bars can suppress jumps, understate volatility, and alter intrabar ranges. They can also create timestamps where no market observation exists.
That distortion propagates into quantitative analysis. Backtests often assume that each candle is a valid tradable state. If an outage is turned into synthetic candles without an explicit flag, the backtest can react to prices that were never observed. Sonar Sciences describes backtest overfitting audits as checks on whether a strategy result is robust to data and model choices rather than an artifact of the testing setup. Unflagged interpolated bars are exactly the kind of hidden data transformation that can change a test outcome while remaining invisible to the researcher. They can alter entry and exit timing, smooth drawdowns, change turnover, and affect any statistic derived from the return path.
This matters for risk adjusted metrics as well. The deflated Sharpe ratio is presented by Sonar Sciences as a way to evaluate whether an observed Sharpe ratio remains credible after accounting for multiple testing and selection effects. If interpolation changes the return distribution, then the inputs to that evaluation change too. Smoother synthetic paths can reduce measured volatility or serially reshape returns, which can in turn affect the reported Sharpe ratio and therefore the deflated Sharpe ratio. The source explains the purpose of the deflated Sharpe ratio as a correction for inflated in sample findings. It does not provide a numerical case study for interpolated candles specifically, but the implication is direct: if the data generating process is altered by synthetic bars, the corrected statistic is being computed on a modified history rather than on purely observed market behavior.
The same issue appears in cross venue studies. Researchers often compare venues for latency, price formation, spread dynamics, and synchronization quality. Sonar Sciences emphasizes that cross venue data quality depends on preserving venue specific structure during alignment and normalization. Flagging interpolated candles helps preserve that structure because it separates observed bars from synthetic placeholders. With a flag, a researcher can exclude affected intervals, downweight them, test sensitivity to their presence, or require that cross venue comparisons use only bars observed on both sides. Without a flag, synthetic bars can masquerade as evidence of real market activity and contaminate lead lag estimates, correlation studies, and venue level anomaly detection.
The practical lesson is not that interpolation is always forbidden. It is that interpolation changes the meaning of the data. A gap filling routine can be useful for maintaining matrix shape, feeding models that require regular spacing, or keeping chart displays continuous. But once synthetic candles are inserted, the dataset is no longer a pure record of observed trades. The interpolation must therefore be explicit and machine readable. A flag at the candle level preserves analytical integrity by letting downstream systems distinguish observed market history from generated values. That is the difference between acknowledging missing data and accidentally rewriting it.
Covered in depth in the Cross-venue market data & signals pillar hub.