You need to compare apples with apples. The point value for the NQ, ES, YM are different but overall the volatility is proportional. 1 tick on the ES is $12.50 and 1 tick on the NQ is $5.00. When the ES moves 1 point, the NQ will move 10 ticks on average. As a rule of thumb both instruments move proportionally that's why they are priced like they are in terms of commission. Unless you take that into consideration, your numbers are out of whack. In other words, no matter the symbols (ES, NQ) it's all the same.
The following 2 users say Thank You to trendisyourfriend for this post:
This post has been selected as an answer to the original posters question
I personally get my best results from combining scalping and catching large moves. For example, there are some minor price congestion areas that are drawn by my software. On CL. when price makes a quick pop through these areas and then pauses, almost 90 to 95% of the time it will do a quick 5 to 10 tick retrace. I just go for 5 ticks with a 6 to 7 tick stop and catch it. I call it a reverse scalp and it's just a quick counter trend move. I can catch quite a few of these during the day. Of course I don't try them at times of high volatility since a big spike can occur at these congestion areas. But from watching price on small timeframes for several years, this is an extremely high probability setup for me to catch.
On the other hand, I also use strategies that allow for 20 to 100 tick moves . So, depending on the market conditions for that particular day, I will use whichever strategy that makes more sense..
I do know for a fact there are some great scalpers out there who do very well. But in most cases, it takes years of screentime to build that kind of proficiency, and for the average newbie...they probably won't make it if they just try to be a scalper/ There's just too much work involved. It's almost like trying to be a top athlete or fighter jet pilot. You need to have a feel or instinct for price movement and volatility and that only comes with time.
Failure is not an option
The following 10 users say Thank You to lancelottrader for this post:
The other issue you raise is one of spending years to become a good scalper. The problem is the competitive edge of the scalper has nearly been totally eroded by highly advanced algorithmic technology. Worse still, the scalper in 5 years will be facing a much more advanced technology making the task even more daunting.
Last edited by djkiwi; November 8th, 2012 at 06:32 AM.
The following 2 users say Thank You to djkiwi for this post:
There's one extremely fantastic thing about scalping strategies - the return distributions are highly Gaussian. This means 3 things are true:
i. The standard deviation of your sample mean varies inversely with the square root of sample size.
ii. 95% of your PnL will fall within 2 standard deviations of your mean.
iii. The derivations of the Kelly criterion are close approximations: one of which states that your compounded rate of return increases with the square of your Sharpe ratio.
A reasonable expected profit for a single trade on a scalping strategy is 0.16 tick per contract after slippage and costs with a standard deviation of 8 ticks. (It means approximately 95% of your PnL is contained within +/- 16 ticks, you might achieve this by having a 50/50 chance of getting 2 ticks with a stop loss of 10-12 ticks and average slippage and commissions coming round to 1.68 ticks per round trip, nothing extremely unimaginable.) By definition of "scalping", your holding periods are significantly shorter than "swing trading". This means you can generally trade more times per day than a "swing trading" strategy. If you scalp 252 days a year, 89 times per day, your strategy has a Sharpe ratio of 0.16*sqrt(252*89)/8 = ~3. What some guys like Saluzzi and Arnuk glorify as "HFT" simply takes this same, simple principle and applies to multiple instruments, yielding perhaps 5000 trades per day. That is enough to get a SR of 22.
You don't need a spreadsheet or Monte Carlo simulation to prove mathematically that a scalping strategy is actually superior.
The real reason to begin with why scalping sounds bad from a R:R analysis is because risk-to-reward ratio ("R:R") measured by stop loss orders is BS. Most of the people aiming for 2 ticks don't actually have a persistent edge of 2 ticks. And most of the people setting stop losses of 10 ticks don't actually "risk" 10 ticks.
Is it possible to scalp for a very long time consistently with a normal distribution with a mean of about +1~2 ticks and standard deviation of 8 ticks? On retail software? Yes. I made a rough, quick port of a strategy from C++ to .NET for NT 7 and it looks like this even without important components, almost normally distributed from start to finish:
Please register on futures.io to view futures trading content such as post attachment(s), image(s), and screenshot(s).
It may seem very, very weird that some trades have >$1000 losses. This is because I don't believe in placing stop loss orders after I have an open position.
"Highly advanced algorithmic technology" does not mean highly advanced trading reasons. Most of algorithmic trading consists of agency-based large order executions, for reasons such as "I just inherited $200 million worth of shares and do not have cash to pay inheritance tax." Gaussian distributions are also very important if your trading involves clients like this, because if your trades are too weird (e.g. you use some hocus-pocus risk management based on R:R and stop losses), no one wants to buy your products or trade with you.
I don't want to detract from the topic, but writing in a forum like this is nice, sometimes better than sitting in Goldman Sachs. I think you make the wrong assumption that making $10 million per year is a reason not to spend time in this forum. For example, I joined only to watch the Nanex webinar, then felt bad about winning one of the giveaways without having posted here before, so I decided to stay on to post once in a while to make up for it. I think it is enjoyable to interact with people regardless of how much money you make, or how much you take back from them.
<I removed this part because it doesn't add value to the discussion.>
Last edited by artemiso; November 10th, 2012 at 11:15 AM.
Reason: Removed the last part
The following 13 users say Thank You to artemiso for this post:
To avoid confusion we need to differentiate between retail scalping and computer scalping as they are vastly different animals.
As computers become more powerful their role is expanding from purely HFT to scalping. They are able to analyze billions of iterations in milli-seconds, assess the probabilities and make the trade before the retail trader can even click the mouse button.
These machines trade with no emotion, are controlled by "seated" bosses who pay no commissions and are highly capitalized.
This means computers have a massive edge over the retail scalper who may be trading with a hangover the night before, or dealing with family tragedy or a myriad of other issues. We saw what happened when Gary Kasporov the long time world chess champion took on deep blue the IBM super computer. He got his head handed to him in the rematch and that was in 1997. Computers have advanced considerably since then.
So computers are using increasingly advanced algorithms to further increase the edge over retail scalpers. This is more of a reason for a retail trader to avoid scalping than embrace it as he is trading against a super advanced computer every single day. So I believe your approach may work for computer scalping but not retail scalping.
2. Retail Scalping
The problem with this approach is retail traders are not well capitalized. I'm not sure how you are doing your position sizing but if you assume the average retail account is $10k and they incur a string of $1,000 losses it will prematurely end their trading careers. Or you could have one catastrophic loss. It is a mathematical certainty this string of losses will happen sooner or later.
Without understanding the mathematics in detail, it seems to me your approach increases the risk considerably to cover the commissions and slippage costs whereas increasing the timeframe and trading less regularly reduces the risk and optimizes profit. I'm not sure how else you can do this if you are working off 50/50 probabilities with a market based stop. As I pointed out in my previous posts most retail scalpers need to improve their odds by 20% just to cover their expenses. I'm not sure we need to use advanced mathematics to understand that the scalper must have a huge edge to cover these costs. Why buy the broker and market maker a new BMW each year?
A trader scalping for 252 days at 89 trades per day is not practical or even mentally possible. Fatigue will be the first guest and then the trader will probably burn out and end up in a straitjacket in a mental asylum. What you describe is working hard but not smart and most likely even a hard core scalper would quickly find trading a chore than a pleasure. On the other hand the swing trader can analyze the trades the night before, place the trade and then go to the beach or even go to work in a 8-5 job. Meanwhile the poor old scalper has to bang out another 88 trades to pay his two new bosses (the broker and market maker).
The Kelly criterion is also highly sensitive to the accuracy of the expectancy. We know retail traders both overestimate and misjudge their edge which causes them to trade larger size, therefore increase their risk and in many cases lose all their money.
Prior to the financial crisis I'd been looking around at different risk management/position sizing models and looked at a Kelly type of approach. To calculate the optimal percentage of my portfolio to put at risk, I worked out the percentage of trades expected to win as well as the return from a winning trade and the ratio performance of winning trades to losing trades. Lucky for me I had many years of data to call upon. The first issue was there had been quite a bit of variability in these numbers over the years and it appeared using the Kelly formula would likely increase the volatility of my returns in the future. It's just as well I didn't use this method because during/after the crisis, those setup probabilities changed significantly which would have had a major negative impact on returns if I'd moved from my simple risk/reward model to a Kelly approach.
The point is markets change and change frequently so any inaccuracy of the inputs into the Kelly formula will be seen in future return volatility.
The beauty of risk/reward measured by stop orders is it's simple but effective. I've used risk reward minimums for a long time and also a simple rule of never risk more than X% on any trade. It's simple, gives me a fair idea of what I'm going to earn in a given year because I know my typical trade risks and is easily measurable.
Once you have some trading experience and incur a string of losses, or one catastrophic loss it can induce all types of emotions. In other words, it's easy to come up with a statistical model and post it in a forum but putting it into operation and trading it live with your own money is an entirely different animal.
Last edited by djkiwi; November 13th, 2012 at 12:33 AM.
The following 6 users say Thank You to djkiwi for this post:
That is interesting. I've read this before. This was written in 2005 however and a lot has changed since then. I'd like to hear his views now given the significant increase in the presence of algos since then. I think this part was very interesting and couldn't agree more.
Q: why don't you have any problems with closing out the position and even taking the opposite direction? shouldn't a trader stick to his opinion?
A: No, definitely not. An analyst or some kind of guru has to stick to it, but a trader should have no opinion. The stronger your opinion, the harder it is to get out of a losing position.
The following 2 users say Thank You to djkiwi for this post: