statistica si probabilitati

[deleted on request 01]

se pricepe cineva la statistica de aici? as avea nevoie de un sfat profesionist si nu de pareri.

[dirzuandreiovidiu]

Ce anume din statistica te-ar interesa, concret? Vrei sa verifici, cumva, o strategie?

Editat August 23, 2011 de dirzuandreiovidiu

[deleted on request 01]

da, as fi vrut sa vad daca statistica imi poate face distinctia intre un drawdown normal si o strategie esuata. am in minte aici exemplul celor de la long term capital si evolutia lor. atasez graficul.

[dirzuandreiovidiu]

Ar fi fost bine daca ai avea de la ei date concrete a acelei pante a profitului. Asa, vizual, e cam greu de verificat a LTC. Din cate am inteles cei de la LTC au verificat istoricul pe ultimii 10 ani si, daca ar fi facut-o pe inca 20 de ani, acest faliment nu s-ar fi intamplat, asa zic specialistii. Apropos, pe ce time frame vrei sa iti faci back-testing-ul?

[deleted on request 01]

timeframe h1 pentru monezi si pentru date istorice din anii 70 am folosit daily s&p si daily coca-cola.

eu am citit despre ltc ca aveau o strategie bazata pe corelatie si in caderea aia corelatia s-a schimbat de la 90% la 70% si strategia nu a mai mers. dar deasemeni mai stiu ca dupa ce au fost salvati au mai functionat 1 an in care au avut profit.

[gigi]

E complicat cu drawdown-ul. Ceva relevant:

 

In the summer of 1997, Taleb predicted that hedge funds like Long Term Capital Management were headed for trouble, because they did not understand this notion of fat tails.

 

Just a year later, L.T.C.M. sold an extraordinary number of options, because its computer models told it that the markets ought to be calming down.

 

And what happened? The Russian government defaulted on its bonds; the markets went crazy; and in a matter of weeks L.T.C.M. was finished.

 

Spitznagel, Taleb's head trader, says that he recently heard one of the former top executives of L.T.C.M. give a lecture in which he defended the gamble that the fund had made.

 

"What he said was, Look, when I drive home every night in the fall I see all these leaves scattered around the base of the trees,?" Spitznagel recounts.

 

"There is a statistical distribution that governs the way they fall, and I can be pretty accurate in figuring out what that distribution is going to be. But one day I came home and the leaves were in little piles. Does that falsify my theory that there are statistical rules governing how leaves fall? No. It was a man-made event."

 

In other words, the Russians, by defaulting on their bonds, did something that they were not supposed to do, a once-in-a-lifetime, rule-breaking event.

 

But this, to Taleb, is just the point: in the markets, unlike in the physical universe, the rules of the game can be changed. Central banks can decide to default on government-backed securities.

 

BTW, graficul tau seamana foarte mult cu asa-numita "distributie Taleb"

 

http://1.bp.blogspot.com/_RIqPsj7KzEY/SNcKoIpzjsI/AAAAAAAAAAk/ASDi47BN-zk/s320/turkeyblowup.png

 

Sute de hedge-fund-uri au un grafic similar. De ce? Pt ca e in interesul lor sa faca asta.

 

Daca vrei sa ai o idee despre bazele back-testing-ului, citeste cartea asta: http://www.amazon.co.../dp/0470128011/ Sunt doar bazele, mai e mult de invatat pe langa.

[deleted on request 01]

am sa citesc cartea.

Editat August 24, 2011 de lamishto

[dirzuandreiovidiu]

@lamistho

 

 

timeframe h1 pentru monezi si pentru date istorice din anii 70 am folosit daily s&p si daily coca-cola.

eu am citit despre ltc ca aveau o strategie bazata pe corelatie si in caderea aia corelatia s-a schimbat de la 90% la 70% si strategia nu a mai mers. dar deasemeni mai stiu ca dupa ce au fost salvati au mai functionat 1 an in care au avut profit.

 

 

Pentru backtesting profesional, daca te intereseaza, iti pot face pe 1h, pentru 98 eur per pereche (sau alt instrument tranzactionabil) si pe daily, pentru 33 eur per pereche (sau alt instrument tranzactionabil). Restul detaliilor le vorbim pe privat sau pe forum, dupa cum doresti. Mentionez ca backtestingul este realizat manual in format xls sau xlsx (Microsoft Office Excell 97-2003 sau Excell 2007).

[deleted on request 01]

nu backtestingul e problema. problema daca te uiti atent la primul post era cum sa deosebim drawdawnul normal de o strategie care nu mai merge. eu am o rezolvare practica dar era bine daca as fi vazut si o rezolvare statistica.

[dirzuandreiovidiu]

Ce intelegi prin drawdown normal? drawdown de 5-10%, sau cat, concret? Pana si drawdownul LTC exista doar ca nu se vede pe candelele alea, el era poate la timeframe saptamanal, ori zilnic, de aceea nu se vede.