To all those who are tempted to outhink the SPY system (for
example, by selling into the current rally or by not taking trades
when the market appears vulnerable to you), I suggest an
experiment. Set up a paper trading account with your broker
and invest the exact amount in both your real account and your
paper account, except that the paper account makes trades exactly
as the system generates them whereas the real account is managed
the way you normally tend to manage it. I suspect that most
people will be surprised at how the paper account will gradually
outpace the live account. At least, most people who are not
extremely adept traders, but then again, these don't need a
mechanical system.
Since you're showing this on the SPY board, I assume that you're
planning on making it available. Could you discuss your plans
for this mechanical system?
Hasn't the system been trading live longer than that, i.e.,
since before it was released? What would be useful to know it its
results since the last time it was tweaked.
There's been some discussion here on backtesting, so I thought I
might clarify the ntion a bit.
In a recent post, Matt said that the SPT system wasn't developed
via backtesting. This is true, however no systems are
developed via backtesting. Systems are tested through
backtesting and frequently refined through optimization. An
example might help.
Let's say I think that trading SRS on a 5 minute chart would be
a good idea. Furthermore, I think that maybe moving average
crossovers would work. So I set up a 5 minute SRS chart with
fast and a slow moving average, say 5/15 EMA. This is my
hypothesis. Then I test it via backtesting. If it looks
profitable, I might then optimize, which means my software tests
various moving average combinations to find the most profitable
one. Notice that I didn't literally develope the system via
backtesting: I tested it that way and refined via optimization.
Of course, in a way the system is developed via backtesting:
in the sense that this process can have many iterations, each
one tested via backtesting, so that, effectively, backtesting is an
integral part of the process.
Now, the problem lies in optimization. Specifically, by
optimizing a system, I run the risk of curve-fitting, which simply
means that although I've found the best possible combination of
variables for a given period there's no guarantee that this
combination will continue to work in the future.
There are many ways to avoid or minimize the effect of
curve-fitting. One is walk-forward analysis: a system
is optimized for a given period and then tested during period
outside the optimization period (out of sample). Another way
is to see if the given system works with instruments other than the
one for which is was optimized. If a system only works for
SRS and not for any other stock or ETF, it is likely that the
system was curve-fitted.
Now, regarding the SPY system, I understand that it was not
developed with a heavy reliance on optimization. According to
Matt and Tom, much of the system was developed using techniques
that Matt has been using for years. Furthermore, it is true
that the system wasn't developed via backtesting, but it was
tested, and presumably refined, via backtesting. Consequently, the
results (90% winning ratio) has been calculated using backtesting
data so can't be taken as indicative of future results. It's
not that the system might stop working in the future; it's possible
that it never worked as well as the data suggests because that data
is culled from backtesting not from live trading.
What would be helpful would be a running record of the system's
performance in reatlime (I understand that Matt has been running
the system live since last November) and, if possible, a test to
see how the system works with instruments other than SPY.
Michael, That's not what I said. I was using an analogy.
My point was that in any probabistic situation, history
doesn't change the odds. As an analogy, I was pointing out
that even if you've tossed 2000 heads in a row, the odds of the
next toss being heads are still 50/50. So, regarding the SPY
system, let's assume that any given trade has a probability to be a
winner or a loser. That probability is not changed because
previous trades worked out in a certain way. It is what it is
based on whatever factors determine it.
Actually, I don't think that it matters. If you toss
thirty heads in a row, the odds of the next toss being tails is the
same as all the other tosses, 50/50.
The system certainly has two out of three losses in a series.
But it didn't happen very often. So, the odds of it
happening aren't very high. And yet it did. Hence, the
anomaly.
1. We're in a statistical anomaly and the system will soon
find its groove.
2. Backtested results won't match live trading results and
we'll have to wait and see what the real strength (or weakness) of
the system is after it's been trading live for a while.
I don't see any reason not to use a tight sell stop for the day.
Should the market reverse and go up then you could sell at
close, thereby preventing the possibility of another gap down
tomorrow.
I used to teach English, so I'm curious about how people read.
Specifically, with the wealth of information on the FAQ and
numerous posts here, how do you come up with "a long term from this
system" when there is clearly no such thing?
True, and I'm following the system regardless of what how my
emotions conspire to undermine my trading.
However, we should be aware that the overwhelming majority of
those "16 years" are backtested, not traded live. From what I
know of the system, it appears to be robust and not the product of
clever optimization, so I expect it to perform well in real
trading.
Still, until that happens, it behooves all of us to be aware of
the difference between backtested and live trading results.
Automation certainly helps, but you wouldn't believe how many
people discuss whether they should "take" a trade or not. The
bottom line is that human psychology can be a formidable foe.
In fact, mechanical systems are rarely mechanical systems.
Most people treat them like multiple-choice tests, taking a
trade if it fits their sense of where the market is going, or it
fulfills some emotional need. The real challenge when running
a mechanical system is adhering to it.
In that regard, I think that it may actually help that the SPY
system isn't free. Perhaps because people are paying for it
they may be more likely to follow it.
Today is a great test. Every emotion generated by my
glands is telling me to get out of the market. The system is
telling me to invest more in it. I did what the system said
and I feel good about that. I'll feel even better if, after a
few more live trades, the system performs in live trading as it did
in backtesting.
Newsletter
Subscribe to our email list for regular free market updates
as well as a chance to get coupons!
The community is delayed by three days for non registered users.
Following the system
Posted by algyros on 5th of Jul 2011 at 06:46 pm
To all those who are tempted to outhink the SPY system (for example, by selling into the current rally or by not taking trades when the market appears vulnerable to you), I suggest an experiment. Set up a paper trading account with your broker and invest the exact amount in both your real account and your paper account, except that the paper account makes trades exactly as the system generates them whereas the real account is managed the way you normally tend to manage it. I suspect that most people will be surprised at how the paper account will gradually outpace the live account. At least, most people who are not extremely adept traders, but then again, these don't need a mechanical system.
A possibility of what: the
SPY system single entry scale out statistics
Posted by algyros on 30th of Jun 2011 at 02:01 pm
A possibility of what: the system going short or of making it available?
Yes, nice trades. Since you're showing
SPY system single entry scale out statistics
Posted by algyros on 30th of Jun 2011 at 01:21 pm
Yes, nice trades.
Since you're showing this on the SPY board, I assume that you're planning on making it available. Could you discuss your plans for this mechanical system?
Thanks for the tip. What
browser speed...
Posted by algyros on 26th of Jun 2011 at 08:51 pm
Thanks for the tip.
What was the anti-malware program?
Hasn't the system been trading
Posted by algyros on 20th of Jun 2011 at 09:59 am
Hasn't the system been trading live longer than that, i.e., since before it was released? What would be useful to know it its results since the last time it was tweaked.
Backtesting
Posted by algyros on 18th of Jun 2011 at 10:36 am
There's been some discussion here on backtesting, so I thought I might clarify the ntion a bit.
In a recent post, Matt said that the SPT system wasn't developed via backtesting. This is true, however no systems are developed via backtesting. Systems are tested through backtesting and frequently refined through optimization. An example might help.
Let's say I think that trading SRS on a 5 minute chart would be a good idea. Furthermore, I think that maybe moving average crossovers would work. So I set up a 5 minute SRS chart with fast and a slow moving average, say 5/15 EMA. This is my hypothesis. Then I test it via backtesting. If it looks profitable, I might then optimize, which means my software tests various moving average combinations to find the most profitable one. Notice that I didn't literally develope the system via backtesting: I tested it that way and refined via optimization. Of course, in a way the system is developed via backtesting: in the sense that this process can have many iterations, each one tested via backtesting, so that, effectively, backtesting is an integral part of the process.
Now, the problem lies in optimization. Specifically, by optimizing a system, I run the risk of curve-fitting, which simply means that although I've found the best possible combination of variables for a given period there's no guarantee that this combination will continue to work in the future.
There are many ways to avoid or minimize the effect of curve-fitting. One is walk-forward analysis: a system is optimized for a given period and then tested during period outside the optimization period (out of sample). Another way is to see if the given system works with instruments other than the one for which is was optimized. If a system only works for SRS and not for any other stock or ETF, it is likely that the system was curve-fitted.
Now, regarding the SPY system, I understand that it was not developed with a heavy reliance on optimization. According to Matt and Tom, much of the system was developed using techniques that Matt has been using for years. Furthermore, it is true that the system wasn't developed via backtesting, but it was tested, and presumably refined, via backtesting. Consequently, the results (90% winning ratio) has been calculated using backtesting data so can't be taken as indicative of future results. It's not that the system might stop working in the future; it's possible that it never worked as well as the data suggests because that data is culled from backtesting not from live trading.
What would be helpful would be a running record of the system's performance in reatlime (I understand that Matt has been running the system live since last November) and, if possible, a test to see how the system works with instruments other than SPY.
Alex
Michael, That's not what I
probabilities
Posted by algyros on 16th of Jun 2011 at 07:46 am
Michael, That's not what I said. I was using an analogy. My point was that in any probabistic situation, history doesn't change the odds. As an analogy, I was pointing out that even if you've tossed 2000 heads in a row, the odds of the next toss being heads are still 50/50. So, regarding the SPY system, let's assume that any given trade has a probability to be a winner or a loser. That probability is not changed because previous trades worked out in a certain way. It is what it is based on whatever factors determine it.
Then every entry will have
SPY system potential long today
Posted by algyros on 15th of Jun 2011 at 03:52 pm
Then every entry will have the same odds. Probabilities don't have memory, so previous trades don't matter.
Also be aware that backtested results aren't the same as live trading performance.
Actually, I don't think that
SPY system potential long today
Posted by algyros on 15th of Jun 2011 at 03:48 pm
Actually, I don't think that it matters. If you toss thirty heads in a row, the odds of the next toss being tails is the same as all the other tosses, 50/50.
Did I miss something? About
market decision
Posted by algyros on 15th of Jun 2011 at 11:30 am
Did I miss something? About what system are you talking?
The system certainly has two
Stock Market 2 Spy System 1
Posted by algyros on 15th of Jun 2011 at 10:07 am
The system certainly has two out of three losses in a series. But it didn't happen very often. So, the odds of it happening aren't very high. And yet it did. Hence, the anomaly.
There are two ways to
Stock Market 2 Spy System 1
Posted by algyros on 15th of Jun 2011 at 09:37 am
There are two ways to explain this.
1. We're in a statistical anomaly and the system will soon find its groove.
2. Backtested results won't match live trading results and we'll have to wait and see what the real strength (or weakness) of the system is after it's been trading live for a while.
I don't see any reason
anyone else having a hard time pulling the sell trigger, doesn't make sense yesterday sell trigger was based on much higher price
Posted by algyros on 15th of Jun 2011 at 09:09 am
I don't see any reason not to use a tight sell stop for the day. Should the market reverse and go up then you could sell at close, thereby preventing the possibility of another gap down tomorrow.
Love to see the chart,
well....
Posted by algyros on 11th of Jun 2011 at 09:11 am
Love to see the chart, but it didn't come through.
Like I said, there are
spy open
Posted by algyros on 10th of Jun 2011 at 10:58 am
Like I said, there are no mechanical systems. They all eventually become multiple choice systems.
I used to teach English,
SPY swing system 3rd entry
Posted by algyros on 8th of Jun 2011 at 03:51 pm
I used to teach English, so I'm curious about how people read. Specifically, with the wealth of information on the FAQ and numerous posts here, how do you come up with "a long term from this system" when there is clearly no such thing?
True, and I'm following the
SPY system scale in 2nd entry
Posted by algyros on 7th of Jun 2011 at 01:05 pm
True, and I'm following the system regardless of what how my emotions conspire to undermine my trading.
However, we should be aware that the overwhelming majority of those "16 years" are backtested, not traded live. From what I know of the system, it appears to be robust and not the product of clever optimization, so I expect it to perform well in real trading.
Still, until that happens, it behooves all of us to be aware of the difference between backtested and live trading results.
Automation certainly helps, but you
SPY system scale in 2nd entry
Posted by algyros on 6th of Jun 2011 at 08:44 pm
Automation certainly helps, but you wouldn't believe how many people discuss whether they should "take" a trade or not. The bottom line is that human psychology can be a formidable foe.
In fact, mechanical systems are
SPY system scale in 2nd entry
Posted by algyros on 6th of Jun 2011 at 04:43 pm
In fact, mechanical systems are rarely mechanical systems. Most people treat them like multiple-choice tests, taking a trade if it fits their sense of where the market is going, or it fulfills some emotional need. The real challenge when running a mechanical system is adhering to it.
In that regard, I think that it may actually help that the SPY system isn't free. Perhaps because people are paying for it they may be more likely to follow it.
Today is a great test. Every emotion generated by my glands is telling me to get out of the market. The system is telling me to invest more in it. I did what the system said and I feel good about that. I'll feel even better if, after a few more live trades, the system performs in live trading as it did in backtesting.