The probability is derived from the likelihood of price moving in a certain direction based on experience or screen time of seeing similar price patterns and technical analysis (patterns, volume, indicators etc.). The other method is time based, wherein if the trade fluctuates between profit and loss for a few minutes or so but is not going in any direction, then I cut the trade. I prefer waiting to re-enter when price moves in my direction than having capital tied up in a position that's not moving anywhere yet.
Reviewing the Results
I started on May 11 and traded 14 days out of 22 possible days in May and skipped two market sessions. There were 108 winning trades, 29 losing, and 2 scratch trades (return of $0). The average losing trade lost $15.96 (median of $11) while the average winner returned $21.83 (median of $16). Anything where the average winner is greater than the average loser is good to see and in this case, about $5 more profit is earned than is lost per trade. Furthermore, the median profits and losses are both more than half as much as the average, which implies the winning and losing trades tend to gravitate to the average. The win rate is 79.14% but it might be biased since there are only 139 trials - as such I expect it to go lower in the future.
The average amount of capital risked per winner is $4,551.76 and per loser $4,071.41. This means that on an average, I tend to risk about 10% more capital on winning positions than I do on losing positions. This is an important observation because risk is almost entirely decided upon before entering the position or in other words I cannot know beforehand the correct amount to risk until after the trade is completed. One explanation is that there is an inherent bias in my method that tends to risk more on winners before they happen however the other explanation is randomness. What I mean by randomness is since there are only 29 losing trades but 108 winning trades, this discrepancy could be explained by a lack of samples. Either the method has an advantage in allocating more risk on winning positions before they happen (without hindsight) or this occurs due to randomness and should fall over time as more losing trades occur. None the less, I am more inclined to conclude that the method is good enough that randomness does not explain the results.
Next is the total gains, less the total losses and commissions which come to a net profit of $1526.30 or an average daily profit of $109.02 for the 14 days. The projected daily profit multiplied by 22 days comes out for a monthly return of $2,398.47 or an hourly wage equivalent of $13.63 assuming a typical office hours of 7 and half hours. For the amount of effort and uncertainty, it appears as though the return is too little to be worth the trouble. However, if the model is shown to be successful for the current level of risk, then theoretically one can increase the risk while maintaining as best as possible the present returns of the model. That $13.63 becomes $27.26 when doubling the risk and now it becomes very worthwhile. If we assume that the statistical returns suffer a 10% degradation when risk is doubled, then it is still worth it.
A reasonable question is why not just quadruple risk right now. The answer is in two parts. First, the transition from the current risk parameters to the next phase (doubling the current risk) must be as smooth as possible for the trader and as a result, the transition cannot be instantaneous. A problem arises in the psychological mindset of the trader because of the effect that money has. Doubling of risk now means that instead of losing an average of $15.96 per losing trade, the average loser will now increase to $31.92. It may not seem significant, however can we reasonably assume that the trader can still perform with the same results while experiencing larger draw downs? What if the trader doubles risk (capital or amount of shares) and now has a doubling in draw downs but only experiences one and a half times the present return? The model degrades as the losing trades now absorb more of the profits than previously.
Consider if I were given a million dollars to take risk with and perform the same results. I'm more than certain that the same results could not be replicated because I am not accommodated or psychologically prepared to deal with draw downs of tens of thousands of dollars which at my current risk level may only represent $15, an amount I am comfortable with losing. What I am advocating is that if the model works with the current risk employed then it should be possible to slowly increase risk while maintaining the same statistical advantage. It may take a year to get to a point where risk can be doubled and have the same results, and maybe another year or two to triple the current risk and have the same level of returns. A tripling represents $40.89 equivalent wage per hour. Easier said than done, but it is none the less a reasonable expectation. The biggest problem is that while risk may increase exponentially, profits per amount risked may only increase linearly.
Correlations and Standard Deviation
Below are lines of best fit for the amount of capital used (risk) per winning trades and losing trades. For winning trades, the majority of returns are distributed towards the lower range. In fact, 85 out of 108 winning trades have a return between $0 to $30 and 92 of the 108 or 85.2% of winning trades used $6000 in capital or less. The correlation between capital used and return for winners is 0.167 meaning that the amount of capital used and the return per capital has no correlation at all. Using more capital or taking on more risk does not seem to suggest greater returns.
One explanation could be that as more risk is taken, confidence in holding the position erodes and as a result I tend to exit prematurely. The key to success is emotional control and discipline which can only occur when the amount risked does not breach a comfort level. Taking on more risk than one is comfortable with can affect logical decision making processes and cause mismanagement of trades. Whether or not that is enough to explain these observations is difficult to ascertain as I would need to track the average amount of time a trade is held. Overall, what this shows is that I need to push my advantage and focus on the strengths. That is lots of smaller returns but with a high probability of occurring and not risking more than $6000 of my capital at any one time. One final observation. Take a look at the outliers for returns less than $40 (above the grey box) which seem to suggest "scared money" where the returns do not justify the amount of risk undertaken.
The correlation for losing trades comes up to 0.51 which is good enough to suggest that more risk means higher losses. From the previous graph we concluded that more risk does not guarantee higher returns while from this graph we conclude that more risk leads to larger losing trades. These two observations alone spell it out very clearly - more risk does not guarantee greater returns and leads to suboptimal returns for the amount of risk taken. More risk is not justified because we are not compensated for the extra risk we take, at least according to this model of course.
Standard Deviation
The standard deviation for winning trades is $18.9 while the average profit per trade was $21.83. For losers, the standard deviation is $19.05 while the average loss per losing trade is $15.96 and so for winners the standard deviation is less than the mean while for losers the opposite is true. Standard deviation measures the dispersion of the profits and losses from their respective means. In this instance, the profits per winners tends to be dispersed below the mean while profits per losers are dispersed above the mean. This is not something we want to see and should find a way to improve it so that the std dev of profits per winner is greater than its mean. An explanation for why this occurs could be that lack of samples and outliers skewing the data. For example, I removed the largest single loss of $82 and the average loss fell to $13.60 from $15.96 (17.35% change) but standard deviation fell to $14.64 from $19.05 or a 30.12% change.
Too Good to Be True? Maybe... Maybe Not
Sometimes, things that are too good to be true usually are because they omit certain key facts that otherwise would not paint such a rosy picture. With everything presented thus far, such as a win rate close to 80% and five times the dollar return for every dollar lost, it seems too good to be true for me as well.There is of course the possibility I'm making it all up, cherry picked the winners and threw out the losers. This would be a big experiment in self delusion and if I did this, rest assured, I'll blow out my account soon enough so for now let's assume this isn't the case. However, I think there are some convincing reasons as to why we can trust the results. Also keep in mind I posted all the trade results.
Reason 1 - No Shorting
My practice account does not allow shorting (betting that price will drop) and as a result I could only buy stocks and hope to sell them at a higher price. This is important because this mini experiment began on May 1st and in the entire month of May there were only 6 days where the S&P 500 closed higher than the open.
Reason 2 - Return of $588.97 Trading Facebook
I traded 3950 shares of Facebook, lost $161 and made $747.97 for a net return of $588. The reason I traded so much Facebook is because of the liquidity (large amount of volume). The graph below shows Facebook stock, and as we can see, price has been falling ever since the May 18 open. The only way I made money was by (for lack of a better word) betting against the prevailing downwards trend. Moreover, IPOs aren't shortable until after some time has passed. So I made profits on a stock that was falling by buying it. What this means is that I was able to pinpoint with a high degree of accuracy places throughout the day where price would reverse against the prevailing trend. I believe that this could not have occurred through randomness alone.
Conclusion
The method seems to work. The only reason it will fail is due to the psychological differences when trading with money you don't care about losing as compared to money you do care about losing. As a result, when I put this into live trading very soon, I will know that any deviations from the model is due to my error in mindset and not the system or technique itself. Doing all this, if for nothing else, has provided me with a key insight of where I will need to focus my energy if I am losing - not the system, but the mindset.
0 comments:
Post a Comment