Hi all, new in the forum which is Great btw. Starting off with a question. Anyone know of a good rescource (book, article e.g.) discussing the characteristics of stock indexes relative their constituents; - effects of changing components -re-weighting -dependencies on sectors -characteristics of different world indexes (not just FTSE NAZ SP) - best ways to replicate index performance using a select number of constituents. Of course it bleeds into qestions of futures versus cash market, arb opps and stuff, but basically what I´m after is a place to find a more profound discussion of the relationship between index behaviour versus the behavour of its components. Appreciate any suggestions Great trading to you all
Certain charting packages allow you to place overlays over the top of a primary chart (i.e. the S&P) In q charts you can look at: $bix.x - S&P 500 banking index 20% of the S&P $hcx.x - S&P 500 health care index 15% of the S&P etc........ You can determine which sectors lead the index You can then combine the more powerful sectors to create your own lead indicator for trading the S&P $BIX.X +$HCX.X +etc... With "Neo-Breadth" software you can create your own Advance/Decline info for the S&P rejecting the NYSE listed Bond funds, REIT funds etc... that generally move counter to the index and thus skew the A/D line. As far as finding a detailed book or report on all the info you are looking for, I doubt you will find that. That type of research leads to an edge which leads to trading profits which leads to secrecy or a very high price tag for the info There is a lot of info on Standard & Poors site http://www.spglobal.com/indexmain500_data.html
Thanks a lot for kind reply and link. As for sector indexes versus SP etc...Yeah, was something like that I was thinking of, finding leading groups of stocks. Europeean markets are at times very correlated, a friend of mine did a clustering study some time ago, found stable relationships between stocks from different sectors and countries, so maybe if you can identify "leading stocks" (lagging/ leading correlation) and combine them into a crossmarket index of your own, then backtest signals. Maybe it could turn profitable versus some of the smaller markets (Dutch, Danish, Swedish). Takes some time to do though, but I gues there are no short-cuts. Looks like Index-Lab at wealth- lab could be a way of simplifying things a bit. I use a statistics package (Statistica) and an old copy of Metastock for charting. Even though research could lead to an edge it usually disappears after some time. I was more thinking or an overview of index characteristics, what particular effects certain composition features has on market behaviour something like Perry Kaufmans exellent "Trading systems" but more specifically on indexes. Found your thread on market profile. Nice. You still use it They? Take care
joe i think i saw an article about the subject in one of the early bnp cooper neff's "insight" - www.bnpcn.com. as far as i remember they claimed that the number of stocks necessary to replicate the sp500 as multiplied during the last 20 years. IMO this is mostly due to extensive basket trading by big investment operations. if you do it on a single market it is most likely not so much a question of how sophisticated the model is, but more a function of speed and commission. this refers to any kind of index arbitrage. the second thing you are referring to (trading indices against each other) seems to be more profitable to me in the first place. though i know now several groups doing this, i still think there is potential. if you combine it with index replication there might be additional edge. i think it might be even more interesting to look at different sector baskets in different countries, since, due to globalization, there is increasing correlation between stocks within the same industry in different countries. always be aware of survivorship bias in your backtesting. this is always an issue, but it is especially important when you try to play with index composition. you can get the index composition from the standard&poors - but they will charge. there were about 300 changes within the composition of the sp500 since the mid nineties - including renaming, merger and bankruptcy. there were 1100 such changes in the sp1500. so this is something that should not be ignored IMO. I assume that using statistica will make it - we use R for things of that kind. for simple charting metastock is not too bad either, in the end you will most likely look at multivariate things within your statistical software. peace
Man, thanks for interest. good points. You know of any standardized techniques for avoiding survivorship bias? Segmenting samples into stable parts inbetween changes? Continuous recalculations of some kind? Results of backtesting are also always approximate and directional rather than specific and trustworthy. As for the replication it is probably true as you say that institutional trading has increased he number of issues needed. Thanks for tip about R - software, is it actually free? Looks very good. To those interested, I actually found a whole bunch of full text articles (on this and other subjects) at sciencedirect.com. Everything from crossmarket correlations, lead-lag relationships , to pit versus screen traders and running stops in fx markets have apparently been subject to academic studies. Hard to chew most of it tough. Also found two authentic articles: read'em and weep. "Darwinian selection does not eliminate irrational traders" European Economic Review Too bad, eh.. "Behaviorly realistic simulations of stock market traders with a soul" Computer Physics Sounds like the old proverb Why do the wicked prosper? _________
survivorshipbias I am very much afraid of it when it comes to long-short plays. within almost all strategies of that kind I try to exploit a very little expectation value - average return per trade will be in most cases just a few cents. thus i am very vulnerable to any kind of distortion, be it trading costs (commission+slippage) or any kind of bias contained within my database. the better I am in building models the more vulnerable I am against these threats, since my technology will sooner or later fit me into that corner of statistical fluke. honestly speaking we did not even try to find work arounds, rather tried to rebuild historical indices and get the data from then. was - and still is - enormous time consuming work. thus I cannot answer your question which is the best proxy to reduce survivorship bias effects. If you find something, maybe you'll post it? thanks for the url - looks interesting at first glance. R: I think its free, I come back to you on this if you wish, but you should find a lot of material on the net. peace