Before we look at some intraday strategies in detail, I have to admit I’m biased. I don’t believe in day-trading for several reasons:
- No basic edge: studies point to merely 5% of day-traders being consistently profitable. In contrast to long-term investing, where a „dumb“ edge exists in participating in economic growth, trading is a Zero Sum Game. The average of all traders must lose by the amount of fees they pay.
- Less sustainable edges: in this environment trading opportunities evolve and go through cycles quickly, so edges become harder to maintain. Long-term risk premia are much more stable, but don’t play a role in an intraday time-frame.
- Noise overshadows signal: advantages that are quite pronounced at longer time horizons will become more random the shorter the time frame becomes.
- Too simple: Patterns that are clear and simple have been known for ages, they should have been arbitraged away a long time ago. Can there really exist such strong, constant behavioral market forces, that some intraday patterns persist even after the advent of algorithmic trading?
- Not robust: on the other hand complex pattern and indicator combinations often rely on over-fitted backtests – optimized parameter sets are unlikely to perform as well in the future as they did in the past.
- No performance advantage: most realistic strategy statistics show no advantage over more long-term strategies which are less time consuming and less error prone. There is a limit to sustainable returns set by the risk one needs to take to get there. These risks can be controlled, but that prevents leveraging up beyond a certain point. Overtrading will lead to extreme volatility and diminishing growth rates. Where are the day-trader billionaires?
- Industry of selling secret strategies: almost everyone in the short-term space tries to sell you stuff, often with very little reliable statistical data. This, to me, is another sign of the difficulty of making money purely from trading. If I had such a great money making machine, I would just procure as much capital as possible to trade it and not bother with selling its signals. If I were convinced of its sustainability, I would give back by making it freely available.
- Sitting in front of the screen watching prices all day compounds emotional mistakes.
- Limited capacity: limits to liquidity become more pronounced in the short-term space. Slippage quickly becomes a decisive cost factor which is hard to incorporate realistically in backtests. This low capacity may exclude many institutional players from using these strategies and give an advantage to smaller retail traders.
- The advantage of a reliable short-term strategy lies in the possibility to bet on your edge more frequently, which could lead to a higher Sharpe ratio (higher returns at lower volatility). The downside to that is that commissions quickly add up and very often eat up all the profits and more…
But perhaps this skepticism serves as healthy protection when looking at the short-term trading space. It prevents me from doing too many dumb things and to take things at face value too easily.
But I’m also curious and I keep my eyes open in my research – the dynamics of the space are quite fascinating.
So when I come across empirical evidence, supported by my own tests, that something seems to work overwhelmingly well, I feel compelled to dig deeper and even to put some money on the line to figure out what may be going on there. Maybe, in the face of empirical evidence, I will even change my mind.
Fading the Opening Gap
One strategy that intrigued me is fading the S&P opening gap using cost-efficient, ultra-liquid futures. It buys / sells at the market open against the morning gap direction and closes the position when price retraces back to yesterday´s closing. Its simplicity and pre-defined time window mitigates some of the problems inherent in intraday strategies.
Very good performance statistics boil down to profits of equal size to losses with a win probability of about 60% to 70% with little optimization (see rules below). This holds up in long-term independent backtests from several sources (eg. here – with lots of strategy detail – and here) as well my own tests on recent data.
This simple strategy is extremely easy to execute manually, with no discretionary judgement necessary and takes just 15 minutes a day with no additional screen time after the open. Betting opportunities arise almost once every trading day, adding up to an estimated number of 100 to 150 trades per year. The strategy is uncorrelated to other portfolio assets and strategies and can easily be added to a combination of investments and strategies requiring only temporary margin.
After fees and slippage a minimum win probability of about 52% is needed to reach a profitable strategy with a reward to risk ratio of 1 : 1, when using cost-efficient index futures. To reach a decent annual return a win probability above 55% would make the strategy worth trading – so 60% to 70% is very good for an intraday strategy. Let’s say that there are 8-10 valid setups a month, then returns would add up to an average of 2% to 3% monthly risking 1% per trade (a fairly conservative risk level with a big margin of safety) – easily beating a buy and hold allocation to the S&P as well as most longer-term strategies.
If this edge is normally distributed over time at a 66% win rate and the average profit equals the average loss including cost, then we can use the Kelly Criterion to calculate an optimal bet size. This results in risking 8% per trade at a quarter Kelly – the full Kelly bet would risk a whopping 32% of your capital per trade.
Optimal bet sizes in percent of capital calculated with the Kelly Criterion in Excel: =((A1/A2)*A3-(1-A3))/(A1/A2). More on bet sizing here.
Say, for example, you want to make an average of $5.000 before taxes monthly – with around 9 trades per month, $22.000 of capital would do. Which, I think, is utterly ridiculous as it would imply a consistent return of 24% monthly – „that’s physically impossible“ comes to my mind.
It is easy to see how such a direct calculation (9 trades per month with 6x 8% profit – 3x 8% loss = 24%) will lead to serious overtrading and a blow up as soon as the edge is not totally consistent. Things like a non-normal distribution of returns or a serial correlation between losses leading to long trends of underperformance would derail it quickly and volatility would be huge, even if your statistics were spot on.
Such unconstrained bet sizing is the source of many of the myths surrounding the use of day-trading as a means to make a living off a very small capital base. It might work for a while, but it is unlikely to be sustainable over any period of time.
Still, the more sensible rule to risk 1% to 2% of capital per trade (a very basic position sizing technique proposed by scores of trading and investment books) will make most phases of underperformance survivable as long as a long-term edge persists. Given the performance stats above this would lead to 2% to 4% monthly profits. Capital of $125.000 to $250.000 would be sufficient to support a $5000 a month lifestyle. As many people buy a house for more money, that certainly seems achievable. When using futures to trade it, the strategy can even be run alongside a fully invested portfolio of strategies as it only uses margin temporarily.
This clashes with the often quoted statistic: „only 5% of day-traders make profits consistently“. Why doesn’t everyone simply exploit such a clear edge, spending just 10 minutes a day? It is certainly not a very difficult strategy.
As it is easily programmable, hordes of algos should take this obvious advantage out of the market in no time, or not?
There must be an overwhelming amount of dumb money pushing prices away from the close at the opening, only to be faded back to the previous close 70% of the time. Does that make sense, if the tendency of the gap can usually be seen hours before the actual open in the continuous futures market?
Where lies the fallacy or why could it continue to work?
- Robustness may be an issue. There is no real reason why the edge may not suddenly vanish or cycle on to a different strategy.
- Doesn’t work across all markets. The best concepts tend to work everywhere eg. value and momentum.
- No basic edge. While over 70% of overnight gaps are filled during the day, the basic return for this strategy is still zero. It needs extra rules to make money. This is quite typical for a mean reversion strategy: the size of profitable trades is defined by the size of the gaps, the losses are unconstrained: rare, but large, losing trades will bring down the profitability of the strategy.
- A more nuanced return profile. In real trading many trades likely will not hit the profit target nor the stop loss, resulting in a lower average profit and loss than the initial risk that was taken on each trade and therefore a lower overall return.
- Popularity. The strategy is well known, but other ideas trading the opening, that often trade in the opposite direction from the same entry point, may be more popular.
- The tendency for gaps to close with a probability greater than 70% has been around for decades.
The universe of strategies using the market open
A good idea is to look very closely at the range of popular strategies that are used by day-traders around the time of the market open. Some currently show an edge, others don’t – it is possible that an edge cycles between these strategies. Some are complimentary, others cancel each other out and each uses variations leading to different outcomes. I concentrate on one additional strategy concept in particular and list some of the other techniques I came across.
The Opening Range Breakout (ORB) is probably the most well known opening strategy due to its cult status since the 1990´s, when it delivered outstanding returns. It follows an opposite concept to the gap fade as it tries to capture a trend that develops when price breaks out of the opening trading range, while the gap trade tries to capitalize on mean reversion occurring right after the open. ORB often enters in the direction of the opening gap right around the price at which the gap fade stops out at a loss. It can also be a continuation of the gap fade, when the closing of the gap occurs right after the open and the move continues in a trend counter to the gap opening triggering a breakout signal.
ORB recently has had a very low win percentage around 35% (low win rates are a hallmark of trend strategies and not a real problem, but I would feel more comfortable with a 40% win rate). From my own observations, I found an extraordinarily high number of false breakouts occurring at several time periods (15 minute, 30 minute and 1 hour opening ranges) – often you will see a false breakout in one direction in the 15 to 30 minute window only to witness another false breakout in the opposite direction during the 1 hour window.
Additionally all intraday trend strategies face a serious conceptual problem: At the heart of trend-following lies the concept of letting your winners run while cutting your losses short. Exiting your positions at the daily close curtails the chance for extraordinary winners that often make up the bulk of a trend strategy’s profits.
Several recent discussions and strategy statistics, I came across, point to the classic ORB as having been an unprofitable strategy after cost in recent years. The strategy is widely available as a commercial strategy package, which is never a good thing.
In fact the opposite approach – fading the ORB – might give an edge. Veteran trader Linda Raschke has said that the mornings tend to be mean reverting and the afternoons trending in nature.
ORB´s popularity may divert attention away from mean reversion approaches like the gap fade.
Additionally many other gap strategies from the 80´s and 90´s trade after price breaks yesterdays close against the gap direction betting on a continuing reversal after the gap closed. This entry point and direction exactly coincide with the exit of the gap fade strategy. (Eg. Connors/Wiliams 10% Oops which recently tests negative).
Gap continuation strategies, similar in concept to ORB, trade in the direction of the gap rather then against the gap – stop runs on these strategies will essentially result in a closing of the gap (a successful gap fade) as stops are traditionally placed on the other side of the gap.
Other variations include:
- Return to open after 5 / 15 / 30 or 60 minutes: essentially an anti-ORB approach that looks quite promising, because of the many failed ORB breakouts. I integrated this idea to some extent in my gap fade setup by trying to get a better execution price a couple of minutes after the open, but hold the position for a complete gap close (see rules below).
- ORB that enters at a predetermined distance from the open (for example calculated as a percentage of the previous day´s range) at any time. Other variations add momentum and trend filters etc..
- Single Stock strategies that often screen for unusually large gaps.
In general studies on single stocks over a wide universe and length of time point to a full gap being more of a mean reversion then a continuation signal. Especially up-gaps led to short term (over the next 1-3 days) underperformance historically.
Other markets: opening strategies work across many futures markets, but in others they fail (which is a red flag). In general these strategies work better on a basket of securities (eg. equity indices) than on single stocks – random volatility equals out between all securities in an index and the signal is clearer. This is found to be true in longer time frames by different authors (eg. Antonacci, Clenow) and should matter even more in lower time frames when random noise increases.
Bringing it all together
An explanation for the validity of a gap fade strategy could be, that it has a smaller following than an ORB or other trend trading strategies around the opening range and therefore takes advantage of anti-ORB arbitrage. If so, it can be expected to be an unsustainable advantage as popularity shifts towards strategy variations that work and cause them to work less well in the future. Much like the ebb and flow we can witness between different investment factor models, for example between value and momentum strategies or short versus long volatility ideas.
Regimes may change between those that favor a trend approach and those that work better for mean reversion.
We could find a mechanism to determine where we are at and / or combine different concepts for trading around the opening for a more robust strategy. Using a Google search, for example, may indicate a popularity shift in progress as many more hits can be found for “fading the opening gap” than for “opening range breakout”…
A pure gap fading strategy seems to be too simplistic to be sustainable over the long run. On the other hand the effect of gaps closing with a probability greater than 70% has been around for decades without any visible deterioration.
Real money test
As the downside is small and the statistics are enticing with little additional effort involved (I run a daily check of my portfolio around the US open anyway), I have decided to run a real money test with a small risk allocation to find out what is behind this. I will keep track of the performance in this blog entry (look for updates here every month or so). When the statistics drop below break even for a period of more than a month, I will switch to observation only.
Rules for fading the gap
To reach the performance described above it is necessary to use screens to identify the gaps that are most likely to close and use stop losses to protect against large trending moves as well as define the initial risk. These ideas are based largely on the excellent book “Understanding Gaps” by Scott Andrews. I only use the two most essential screens to avoid over-optimization as much as possible and add some details of my own around entries and exits.
- Gap size < 40% ATR(20) (ATR = Average True Range over the last 20 days; average distance between daily highs and lows). Small gaps have a higher probability to close than large gaps. In May 2018 around 10 points in the S&P is optimal – this will change with different index and volatility levels which is why I use the ATR to automatically adjust for these changes. If the gap is very small (< 20% ATR) I look for a better price (more then 20% ATR – 8 points currently – from the close) in the first minutes after the open, otherwise the stop loss will be too close to the entry given the current volatility.
- Inside gap: the opening price should be within the previous day´s range. The key is to avoid violent reversals where price gaps beyond the previous candle on the opposite side of the close (above yesterday’s high after down days or below yesterday’s low after up days).
- Entry at open. Often the entry price an hour or two before or right after the open is better than the actual opening price and I put in limit orders to catch a good execution, usually in two lots. A better entry price results in a wider stop loss, which often makes the difference between success and failure as the price runs in the direction of the gap quite regularly before reversing (this would essentially be a failed ORB or gap continuation breakout).
- Position size: my maximum risk is 1% of capital (NAV – in a smaller test account for now). The distance between entry price and yesterday’s close determines the number of contacts traded: ES (S&P mini future) contracts = (NAV/(gap size in points*50))/100
- A Stop Loss is placed at the same distance as the profit exit: entry price minus yesterday’s close.
- Exit to take profits at yesterday’s close; this gives a reward to risk ratio of 1 : 1. I usually split my order in two lots and close the second part 2 or 3 points beyond yesterday’s close as many gap closes overshoot their target a little bit. I put in an automatic OCO order for stop and exit and don’t monitor the price after the entry.
- Time exit: I close the trade after 1 PM as I think the dynamic of the gap close is over by then and we essentially face an equal probability which direction the afternoon price action will take. An exit at the daily close probably yields similar results, but I prefer to have the position open for a limited time only.
Even though the strategy is as easy as can be, in real life trades are sometimes missed and discretionary elements creep in. Take responsibility, if you use these rules and expect results to differ from mine sometimes.
Real money performance
Performance after 4 weeks (4/12/18 to 5/9/18):
The left column shows the actual average profit and loss including fees and slippage (as a percentage of the capital at risk = 1% NAV) and the win probability as well the resulting Kelly-optimal bet size. On the right is a strategy benchmark derived from different backtest sources for comparison.
You can see that the average profit and loss is much lower than 100% of capital at risk, because often the time exit causes results to be somewhere around neutral. That’s something we have to either live with (if we cannot tolerate higher maximum losses at the stop level) or we could use the average loss as our actual average risk and raise position sizes accordingly.
Win probability also comes in below 60% in the first month, but at this point each trade still makes a huge difference to the performance stats with only 19 completed trades to date.
Extrapolated to an annual performance the numbers amount to: 228 trades (the number is likely to be lower as a trade was taken almost every day).
Expected profit per trade: 0,12% NAV (or 12 cents per $1 risked); expected annual return (not compounded as many traders routinely take profits out of their account): 27%; with a more realistic number of 150 trades annually: 18%.
Pretty good – just about „warrenbuffetty“ – for the relatively safe level of risk taken, but a fair bit lower than the results I extrapolated from external sources in the beginning of the article. For now it is a worthwhile strategy to include in a portfolio of strategies using few extra resources.
The Gap Fade Strategy in a portfolio context
As usual I look at all strategies in the context of the overall portfolio to determine my allocation and to avoid duplication by using strategies that have a very similar return profile.
Even if the actual return of the strategy should turn out to be zero, if it is uncorrelated or negatively correlated to the rest of the portfolio, it will still boost the overall risk-adjusted return by dampening portfolio volatility.
A short gap fade has the side effect of hedging any overnight gains of the short volatility and long equity portion of my portfolio (which feels great), but a long position has the opposite effect. I can easily scale a short position to equal my portfolio´s long exposure, which means my overall NAV won’t change materially, when the trade fails. The range between the take profit and the stop loss points will essentially be risk neutral from a portfolio perspective – locking in those overnight gains.
Therefore I prefer the short gap fade in a long biased portfolio even though the probabilities for a long gap fill are slightly higher due to the market´s upward bias. For the test I have decided to take equal position sizes on both the long and short side.
Overall the strategy should be uncorrelated to the rest of the portfolio as long as gap directions are distributed quite evenly – overweighting the short side would lead to a negative correlation which is a very desirable quality. I expect to be able to add the return from this strategy to my portfolio without raising the overall level of volatility by much.
Another psychological advantage for me is, that short-term strategies are exciting while diligent execution of longer term strategies can be (and should be) quite boring. This boredom may mislead one to take action when none is warranted, eg. overriding a trend model by taking profits early. Distraction from long-term strategies while staying involved with the current market situation is very hard to achieve. It can lead to vastly better outcomes when following a good strategy to the letter – if it is watched diligently, but hovering in the background of ones attention.