11.2 The Weekend Dip, part 2

0. Introduction

In the last post, I introduced the "Weekend Dip trading strategy" as explained by Stephen Gornick, and charted the outcome of following such a strategy. In this post I want to describe the outcome in a little more detail. Remember that I had chosen a "buy and hold" investment strategy and a "trade every weekend" strategy as controls against which I could assess the Weekend dip strategy.

All times, days and dates are UTC.


1. Idealised trades
The following charts show the idealised percent gain per trade and the idealised cumulative gain per trade, from selling 1 coin on Friday and buying one coin on the lowest VWAP of Saturday and Sunday. This is idealised because:

• I'm not considering exchange fees
• I'm assuming a trader know ahead of time when on the weekend the lowest point will be and you make the trade at that time.
• People don't generally trade in this manner and are more likely to find a lower price than the VWAP for that day

So it's probable that a real trader will do differently, and possibly better than indicated. This is why I used a control comparison - to reduce the importance of this variable.

2. Cumulative US$ return

The chart below plots the cumulative return in US$ of the "Weekend Dip" strategy (if the past weeks' high is lower than the previous weeks' high, sell one btc on Friday, buy one btc the weekend) and for the controls (sell one btc on Friday buy one btc the weekend; and buy and hold one btc).

This doesn't look too promising. Both the control trading strategy (trade every Friday and weekend) and a buy and hold strategy beat the Weekend Dip strategy in terms of US$ accumulated over the history of trading, although it does earn more than buy and hold at times.
There are a few points, however:

• The Weekend Dip strategy results in fewer trades, so we would expect overall gains to be less - the cumulative mean US$ return per trade and the percentage return for a time period might be better indicators of the strategy.
• It is clear that using decreasing successive weekly highs does not always indicate a down trend - there are a number of trades that occurred near local peaks in BTC value where the high before the peak was at greater than the high at the peak, and on these occasions the strategy resulted in a negative return. A better trend indicator might be useful.

3. Mean US$ return per trade
In this case we cannot compare the Weekend Dip and control strategy to BTC buy and hold. The chart below shows a comparison between the cumulative mean US$ return per trade for the two weekend strategies.

A little better, but still not as good a result as simply trading every weekend. Weekend Dip returns an average of US$0.13 per trade, compared with US$0.15 per trade for the control. On a bright note, on average the Weekend Dip strategy performs better than the control when the trend is flat, or slightly positive. If the Weekend Dip strategy is to perform better than trading every weekend, a better trend indicator is required.

An improve method of assessment is also necessary. The control value is not necessarily the median value, and there is no scale against which a strategy can be judged, other than the control strategy of trading each weekend. Percentage returns would also be useful.

Finally, more realistic simulation will provide a more accurate assessment.

4.  A better strategy
A series of decreasing weekly highs are an easy way to determine if the price is trending. Unfortunately, highs are not always a good indicator of a trending price - the extreme ends of a quantile often show significant variance, and this is also the case here. The median price rather than the extreme price is a much better, lower variance indicator of the weekly price.

Better still is the volume weighted average price (VWAP). This average weights each trade by the size of the trade, and is a nice stable indicator of the weekly trading price. It's less determined by small abnormal trades but by the larger and more serious trades.

If you must decide whether a trend in the price has occurred, the VWAP should be the preferred indicator. The median price is almost as useful. High percentile prices (for example maxima or minima) are not quite as useful in this context.

The new trend indicator therefore will be the VWAP:

• If the weekly VWAP on Thursday morning is below the weekly VWAP at the previous Thursday morning, then sell on Friday and buy on the weekend.

5. A more realistic simulation
There are many aspects of trading that cannot be simply reproduced in simulation. For example, good traders will often mention having a "feel" for the state of an exchange than others, and tend to buy when lower than the daily VWAP and sell when the price is higher than the daily VWAP. I cannot reproduce this effect accurately, or any of many other ways this analysis differs from reality

I can however lose the assumption of selling on the minima of the weekend daily VWAPs. This is clearly impossible since a trader cannot know ahead of time exactly when a market minima or daily VWAP will be reached. Instead, I chose the day which had lower median of lows, which happens to be on Sundays (UTC).

So the strategy becomes:

• If the weekly VWAP on Thursday morning is below the weekly VWAP at the previous Thursday morning, then sell on Friday and buy on the Sunday.

6.  A better scale
Comparing average returns for a "Weekend Dip" strategy to the control of trading each weekend is useful, but doesn't provide a scale to indicate how much better or worse than the control a trading strategy is. If the cumulative earnings from selling on a Friday and buying on a Sunday have a low variance, then slightly better than the control may be a great improvement, and if the cumulative earnings have a high variance, a large improvement may be random chance.

In order to provide a scale of sorts, I simulated buying on Friday and selling on Sunday for on sixty-four random weeks (the same number of weeks the original strategy signalled trading) and calculated percentiles after one hundred thousand replications.


7. Better returns, and a better measure of returns
The chart below is the result of the improvements in points 4, 5 and 6. The VWAP strategy manages a little better than the high strategy, managing US$ 0.11 per trade as opposed to US$ 0.07 per trade for the high strategy. This difference appears mostly due to avoiding trading on Sundays with a higher VWAP than the preceeding Friday, so the strategy relying on differences in suceeding weekly VWAPs seems to be a better trend detector than the original strategy.

However, both are within the middle 50% band of the random trade results - sixty-four random trades from mid 2010 to the end of 2012 beat the high strategy 50% of the time and the VWAP strategy 30% of the time.

7. Quarterly returns compared to "buy and hold".

The quarterly returns for the "Weekend Dip" trading strategies are not completely dissimilar. The percentage returns for the VWAP trend detector are worse earlier on and better later on than the high trend detector. This is probably why the VWAP did better over all - compared to high trend detection, losses were made when the exchange rate was lower and gains when the exchange rate was higher. I think in general using the weekly VWAP to detect trends does better than weekly highs when the long term trend is more clearly up or down, and seems to do worse when there is no trend.

The second chart of the pair are the "buy and hold" quarterly returns. These are hard to compare with the "Weekend Dip" strategy since the gains and losses are on a completely different scale, so I've provided a table of returns below the charts.

The most obvious thing from the table is that there is much more variance in the buy and hold returns, and for last two quarters of 2011 and the first quarter of 2012 (when the market was mostly in a down  trend), the strategy did significantly better than the buy and hold. When the market was either not trending or trending upward, the "Weekend Dip" strategy does worse than a "buy and hold".


8. Summary
I had wanted to show a few more comparisons to non-Weekend Dip trading methods, and also a couple of trend detection methods that are more profitable than either weekly highs or weekly VWAPs, but I've run out of time. However, I think that the foregoing is sufficient to show:

• On the limited dataset available, selling on a friday and buying on the weekend is a valid way to short BTC in a down trending market, with better returns than "buy and hold".
• In an upward trending or flat market, the returns are somewhat hit or miss, and "buy and hold" returns are better.
• Quarterly returns for the Weekend Dip strategy are lower variance than for "buy and hold".
• The "Weekend Dip" strategy does only marginally better than trading on random Friday/weekend pairs, but a better trend detection strategy may provide better gains.
• An experienced trader may obtain much better returns from this borderline trading strategy. I'd like to hear some anecdotal evidence from those who have done well using it over a long period of time.

Donations help give me the time to analyse bitcoin mining related issues and post the results. If you enjoy or find them helpful, please consider a small bitcoin donation:

12QxPHEuxDrs7mCyGSx1iVSozTwtquDB3r

Thanks to the following for use of their data:

bitcoincharts.com: 1kqHKEYYC8CQPxyV53nCju4Lk2ufpQqA2