Forums  > Risk Management  > Looking for a variable which describes abnormal price increases or decreases of a given security  
     
Page 1 of 1
Display using:  

Paul1988


Total Posts: 30
Joined: Oct 2013
 
Posted: 2014-08-20 14:28
Hi,

I am looking for a variable which measures for example the slope of a last price of a given security. The objective is to detect when there are extreme price movements in order to apply certain risk management techniques.

best regards,

svisstack


Total Posts: 290
Joined: Feb 2014
 
Posted: 2014-08-20 14:40
rate of change in some time?

Time well wasted.

Paul1988


Total Posts: 30
Joined: Oct 2013
 
Posted: 2014-08-20 17:01
thanks svisstack. Yes ROC would be an idea. What do you think if i would use Last_Trade Price and benchmark it against a historical 30 days moving average and once Last_Trade is above that benchmark it would trigger my alert. Would it be correct to do it that way?

svisstack


Total Posts: 290
Joined: Feb 2014
 
Posted: 2014-08-20 17:18
dont know what you want accomplish, alert is ambiguous.
when you want trigger alert on 30-days ma crossing; that will work;-)

Time well wasted.

ricko


Total Posts: 73
Joined: Apr 2010
 
Posted: 2014-08-23 13:30
stdev of returns. If current return > n x stdev you can consider that abnormal in that context.

silverside


Total Posts: 1408
Joined: Jun 2004
 
Posted: 2014-08-23 13:36
I would second that, perhaps with an EWMA weighting of the volatility.

radikal


Total Posts: 253
Joined: Dec 2012
 
Posted: 2014-08-23 22:36
ewma volatility + stdev of abs return is probably pretty effective...

I've always done this via simple random walk vol and just mcmc to estimate the percentiles but it's perhaps overkill.

There are no surprising facts, only models that are surprised by facts

Nonius
Founding Member
Nonius Unbound
Total Posts: 12666
Joined: Mar 2004
 
Posted: 2014-08-24 01:48
how about a dynamic AR1 in which the coefficient suggests a shift towards explosive, i.e., not 0, not negative, but statistically significantly positive?

Chiral is Tyler Durden

svisstack


Total Posts: 290
Joined: Feb 2014
 
Posted: 2014-08-24 12:48
I think correct answer depends on what Paul1988 want achieve,

if alert must be triggered when something wrong is with specific stock then measures must be covering all market to compare eg. volatility etc.
In high volatile days it will be not alert if stock have high volatility, also when market is flat then threshold will be much lower than if using statistics based on measure only of this single stock.

also it's risk management so it depends on alert itself, it probably must be triggered as soon as possible, so it will be calculated on irregularly time spaced data or not?

Time well wasted.

Paul1988


Total Posts: 30
Joined: Oct 2013
 
Posted: 2014-08-25 18:39
thanks for all the great answers i will defenitly consider. I am currently building an automatic market making algo which gives certain spreads to the market. Now i am looking for an indicator which describes economic shocks upside or downside for a certain equity in order to adjust the spreads i will send out to the local market. In general there is a norm that i have to give points which reflect a 3% spread of (LAST_PRICE) but i can adjust these percentages however i would like to. In a market which behaves average i would give 1.5% points above and below but i am thinking to adjust these distribution for bullish and bearish markets for a certain equity. I was thinking about taking the slope of the last price into account for a certain time period. So for example if there is a sudden sharp decline in the last Price compared to an observed period of maybe 30 days it would trigger such a re-distribution of spreads, which in turn would minimize my positions as i am giving points to the market which are counter attractive. I hope i could make myself clear.

silverside


Total Posts: 1408
Joined: Jun 2004
 
Posted: 2014-08-25 21:57
There are some common strategies used to mislead bots and it would be worth your while investigating them in detail. In particular I suggest to look at the investigation / trial into the guys who picked off a Scandi MM (I think it was in Norway).

svisstack


Total Posts: 290
Joined: Feb 2014
 
Posted: 2014-08-25 23:00
@silverside: can't find info about that, it's possible to provide more keywords;-)?

Time well wasted.

Azx


Total Posts: 25
Joined: Sep 2009
 
Posted: 2014-08-25 23:59
Here you go, basically two Norwegian day-traders faked buying interest to make Timber Hills algo raise its bids and then reversed the process at a profit. There was a similar case in Sweden where the trader was convicted.

unsmt


Total Posts: 196
Joined: Jul 2014
 
Posted: 2014-08-26 03:42
It looks like you are trying to develop a model to predict a jump of a security. Though it might be not a jump. On the other hand you might be interested deal with market implied forward rates to get perception on future.
You should specify what you actually wish in a general model of a stock price.

svisstack


Total Posts: 290
Joined: Feb 2014
 
Posted: 2014-08-26 14:28
I think he don't want predict a jump, also probably its not possible without having information based on what informed traders decided to pick orders or have information about some publication time of information.

He just want update some values after a jump if I understanding correctly

Time well wasted.

Paul1988


Total Posts: 30
Joined: Oct 2013
 
Posted: 2014-08-26 15:45
interesting Azx, I will have a look at it.

"He just want update some values after a jump if I understanding correctly"

and i already got enough answers here to build something. I will go with the ROC and some slope benchmark.

il_vitorio


Total Posts: 103
Joined: Aug 2014
 
Posted: 2015-03-12 18:33
Thinking a little bit in calculus, a "jump" will drive you your first derivative to an infinite value, so I think the ROC will be the perfect detector.

One thing that I would say is to take care of the time window, one problem that could arise is that you define to close your window, that will give you a pretty sensible stat or to narrow will pass you jumps that were meaningful.

Regards,

il_vitorio

One of my most productive days was throwing away 1000 lines of code.
Previous Thread :: Next Thread 
Page 1 of 1