Is The Opening Price Correct?
The phenomenon known as the overnight effect or overnight drift is a significant topic in quantitative finance, referring to the tendency of asset returns during non-trading hours to exhibit substantial patterns.
This drift has been well-documented across numerous asset classes, including individual stocks, equity indices, cryptocurrencies, and high-yield ETFs.
The question is: when you are backtesting, can you trust the opening price? Whe
Quantpedia has an interesting article called Dangers of Relying on OHLC Prices – the Case of Overnight Drift in GDX ETF. They compared the opening price with the price at 0931 – one minute after trading starts.
Quantpedia’s research extended this understanding of overnight strategies to the commodities asset class, focusing first on the GLD ETF (Gold) and then the GDX ETF (VanEck Gold Miners ETF). They also looked at SPY.
The conclusion is this: You can trust the opening price to be reasonably accurate for SPY, but it is most likely not accurate for GDX.
The Gold ETF Performance
Analysis of the Gold ETF (GLD) revealed that a significant portion of GLD’s performance occurs overnight, while its intraday performance over the last 20 years is negligible. These findings aligned with price action observed in other previously studied asset classes.
The Gold Miners ETF (GDX) bridges equity and commodity markets. In theory, this convergent asset offered the potential for combined overnight drift effects and higher profits.
The GDX Anomaly: Too Good To Be True
When the same analysis procedure was applied to GDX using OHLC (open, high, low, close) data, the results were initially extraordinary. The OHLC data suggested an unusually strong overnight return exceeding 30% annualized.
Specifically, the analysis revealed a significant overnight drift of approximately 30% per annum (p.a.), contrasted by a substantial negative intraday drift of about -25% per annum. Theoretically, this suggested a highly profitable trading strategy: iteratively purchasing at market open and shorting at market close.
However, experience suggested that these results were unrealistic and “too good to be true”. The underlying problem is typically hidden within the opening prices of OHLC datasets.
The Pitfall of OHLC Open Prices
The core issue lies in how the open price is reported in OHLC data: it is derived from the first trade rather than the MOO (Market-on-Open) auction results. This leads to significant discrepancies between anticipated and actual opening prices, making it impractical to expect execution at the reported open prices.
Furthermore, the volumes traded at these initial reported prices must be minuscule, meaning investors are unable to even closely approach getting fills in that price region.
In contrast, closing prices (Close) are usually achievable in reality, often aligning with MOC (Market-on-Close) auction prices on platforms like Yahoo Finance.
Validating Execution Assumptions
To conduct robust methodological scrutiny, the Quantpedia team undertook the essential next step of verifying the anomaly using better data granularity, specifically transitioning to the QuantConnect environment for intraday data.
For the GDX trading strategy (Buy MOC, Sell MOO), the backtest results using the OHLC data exhibited “highly unrealistic, extreme numbers,” approximately 30% per annum. To test the reliability of the open price execution, a robustness test was performed by adjusting the theoretical sell signal execution timing from the Market-on-Open (MOO) to 9:31 AM.
When comparing this approach to the SPDR S&P 500 ETF (SPY), the overnight effect in SPY was confirmed to be “well and alive,” with only a slight decrease in performance when the sell order was executed at 9:31 AM.
The Realistic GDX Performance
On the other hand, the GDX results changed dramatically when the sell signal was executed at 9:31 AM.
While the analysis using OHLC data showed an absolutely unrealistic performance of approximately 30% p.a., Scenario 2 (selling at 9:31 AM) yielded a significantly more realistic outcome.
The GDX overnight strategy performance (before fees and slippage) dropped to 8.58% per annum. This decrease confirmed that the extreme return derived from the naive OHLC analysis was partially a data artifact. Although the overnight drift in GDX prices is definitely present, its magnitude is nowhere near what the OHLC data hinted.
Conclusion: The Need for Robustness Testing
This investigation highlights the dangers of blindly trusting OHLC open prices. The discrepancy between the theoretically derived and practically executable prices underscores potential pitfalls in the naive application of OHLC data.
Researchers and traders must exercise rigorous attention when developing strategies that presume execution at open prices. It is highly advisable to conduct robustness tests and verify performance with intraday execution prices, such as those at 9:31 AM, to ensure more reliable and practically executable results.