Uncovering Hidden Biases In Gold Futures Market

The concept of bias, or recurring price behavior over time, is one of the simplest triggers on which a trading strategy can be built. A bias is an inefficiency of a market or financial instrument that recurs systematically, for example, in specific time frames or days of the week. When it occurs over longer periods, it is usually referred to as seasonality.

In this article, we’ll see how it’s possible to simplify the work of identifying and analyzing these inefficiencies using Bias FinderTM (a software program developed by Unger Academy). At the end of the test, we’ll validate the information obtained with an example so that the rules can be put into practice and independently verified.

 Figure 1

BIAS FINDER

As shown in Figure 1, Bias Finder contains a historical database of the most important futures from different sectors (stock indexes, energy, metals, currencies, etc.), whose data can be analyzed on candlesticks with a duration of 5 to 60 minutes. After selecting the instrument and the time frame, you can evaluate different time horizons: Intraday, Weekly, Monthly, or Yearly. It’s also possible to display the average trend of traded volumes during a day (Intraday Volume). In some markets, this feature can be very useful to decide when to trade in order to avoid moments of low liquidity.

The trend of the selected futures – expressed as an average monetary swing over time – can be calculated in absolute terms and percentages by ticking the corresponding flag. In the Plot Panel, you can decide whether to display the total chart of the relevant period (“Include Total”), the charts of individual years (“Plot all years available”), or the charts of one or more periods, which can be customized as desired.

BIAS ON GOLD

Let‘s now test the potential of the Bias Finder with an example for the CME-listed Gold future (GC), selecting the 60-minute time frame. The data period available for the GC ranges from 01/01/2008 to 30/06/2022. By analyzing the trend of the average monetary swings of the instrument, we can assess whether there’s a recurring price behavior at different time horizons.

Figure 2

For example, considering the GC's daily (Intraday) horizon, Figure 3 shows that after opening at 6:00 pm (exchange time), average prices tend to rise until about 2:00 am, then fall and reach a low around 10:00 am. The average monetary swing between the 2:00 am high and the 10:00 low is about $60, which means that if we had designed a strategy that exploited this bias, we could have expected an average daily gain of about $60 over the period from January 2008 to December 2021.

Figure 3

So, this seems like a good starting point, but it’s important to remember that this is an average value calculated over a period of almost 14 years, so price trends haven’t necessarily been constant over time. To verify this, it is possible to compare shorter historical periods. In Figure 4, for example, the average trends in each year are compared (excluding 2022 to avoid fractions of a year).

We can immediately see that 2013 was a particularly "lucky" year, so much so that it can be considered an "outlier." Furthermore, this inefficiency seems to have almost disappeared in the 2016-2020 period, only to reappear in 2021.

In a few steps, it was possible to assess the persistence of the bias and get an idea of its current state compared to the past and its potential future evolution. Using conventional methods, this would undoubtedly have required more time and effort.

The analysis done here on an intraday basis can just as easily be performed on a weekly, monthly, or yearly basis, thus allowing the evaluation of any seasonality.

Average Monetary Excursion Trend

Figure 4

Another option offered by the Bias Finder is to evaluate the average trend of volumes traded during a day, which, as seen in the case of the GC in Figure 5, makes it possible to identify at a glance the moments of greatest market liquidity and perhaps avoid those in which low volatility could lead to higher slippage in order execution.

Figure 5

STRATEGY TEST

A simple strategy can be coded on a 60-minute time frame to verify what was highlighted in the Gold intraday analysis, making long entries at 10:00 am and short entries at 2:00 am (exchange times). Using a stop loss is recommended, which protects the capital to some extent from excessive losses. After initial optimization, an area of stability between $1800 and $2300 stop loss has emerged. Any stop loss chosen in this range might be acceptable, but looking more closely, the best range for stability is around $2000. Therefore, as a further refinement, it was decided to set the stop loss at $1900.

Figure 6

In Figure 6, we can see the metrics obtained from the strategy, which confirm its performance and its deterioration over time, predicted by the analysis previously made by the Bias Finder. Indeed, we can see that the equity curve flattens from the end of 2013, confirming that this inefficiency persists but has lost some of its effectiveness in subsequent years.

In any case, the backtest results of the strategy are encouraging but not enough to make the system useful for live trading. The average trade is relatively low to cover commission costs and slippage. On the other hand, the large number of trades (over 6400) leaves room for the possibility of applying filters on the entries, eliminating the less profitable trades, which would benefit the overall metrics.

For example, one may consider allowing market entry only when a particular price pattern occurs in order to trade only on the days when the conditions are most favorable for the movement we’re trying to exploit. To select the price pattern, an optimization is performed on a proprietary list of patterns (numbered from 1 to 55) that includes different situations to assess which of these patterns produces better results than the others.

From the results summarized in Figure 7, sorted by average trade, it appears that pattern 12 is the best compromise to simultaneously achieve an excellent net profit. This pattern represents a situation where the body (open-close) of a candlestick formed in the last 5 days is less than 75% of the range (high-low) of the same period.

This pattern definitely adds value and quality to the strategy. While the total profit is reduced from $418,870 (in the case without a filter: pattern 55) to $335,970 (pattern 12), there is a significant reduction in the drawdown, and we go from 6463 trades to 3185, resulting in a reduction of about 50% of the trades, which means an increase in the average trade to 105$.

Figure 7

CONCLUSION ON GOLD BIAS

In summary, bias is one of the easiest market inefficiencies to identify and exploit for a systematic strategy. At the same time, however, it can be a less reliable trigger than others because it can lose effectiveness over time as more players discover and exploit it. To avoid nasty surprises, it’s undoubtedly necessary to regularly monitor its persistence with a numerical and quantitative approach and to have quantitative and objective criteria to decide when it’s time to live trade a system or put it on hold.

Until next time, happy trading!

Andrea Unger