wilmott.com

i am starting a new project with the target to forecast macroeconomics like stocks, interest rates and credit spreads over a time horizon of 12 to 36 months. the thing starts at scratch and first step is to identify necessary data, consequently possible data sources. then we will do some intense analytics of the data as such, meaning that i want to kow and understand my data. then comes the modelling part and i am still unsure how we will address the various problems involved. most likely initial technique will be a factor model optimise-walkforward approach, mainly because it is a straight forward and relatively simple thing. i am a little hesitant to plan for pattern recognition or that kind of thing, since i will have to deal with a relatively small set of data. quarterly GDP for example adds up to only 200 points over fifty years. knowing there were at max a dozen more or less identifiable credit cycles reduces the data further more. so i prefer intuitively clear models to complexity. initially i want to start with few time series and let the project unfold to more complexity. currently i identified the following initial issues of interest.STOCKSdow jones index (i guess i wont find a longer tracked stock index)dividend yieldINTEREST RATESfive interest rate buckets; overnight, two yr, five yr, ten yr, thirty yrcredit spreads for aaa and bbb corporate bondsPRIVATE SECTORprivate savings rateprivate debthousing startsunemploymentPUBLIC SECTORpublic debtgdpbalance of accountbalance of tradeEXCHANGE RATES(an issue since euro is young and proxys - i think - do not go back further than seventies)INDUSTRYproductivityyindustry utilisationpossible data sources:http://www.federalreserve.gov/releases/ ... mlcomments and experience with such projects appreciated. BTW this is not academic. peace

Can we assume that you want a potential profit from this macro modeling? Quite a few people have gone down this road before. You might want to prowl around the macro models being used by the Fed, the BoE, UK Treasury, and Global Insight, although getting through to the actual models may be difficult, especially for GI which requires a subscription. Most of the banks use the GI model, or their own customized version of it.

well, yes this is a profitProject, so i want to make money with it. is here material available on the GI project? (and ... yes, i will do a google search on my own as well ) thanks for the hints. this is what i am looking for.

By "GI" I meant Global Insight. There may be, and hopefully will be, a set of parsimonious equations that you can derive that will facilitate a strategy you want to pursue. Because generalized macro models have been out of favor for so long, except for planning purposes, there may be a chance to prove that they are valid, at least for some sectors. People are always making better products that have been around for decades. Why not? But narrow your intended focus or, if this is a one person effort, you will drown in data though.

good news that it is out of favour . in the beginning this is a one man thing, but i have additional resources. it is just the case that this topic is so vast that i want to define the route before i delegate. i am a little cautious with too complex models in the early stage, because i am afraid that complexity tends to curve fit here very quickly. and as i said earlier even if you go back for decades, you have relatively few data points.

have a look at tramo/seats methods.tramo and seats are especially usefull with montly data and can forecast 24 months ahead with confidence interval.you can find free public implementations : google Demetra and TSW.Once you have forecasts, you want to statistically test whether these forecasts are accurate enough for trading purpose or not. In my experience, it is difficult to answer.

how do you factor in the recalibration of economic figures?

QuoteOriginally posted by: arnaohave a look at tramo/seats methods.tramo and seats are especially usefull with montly data and can forecast 24 months ahead with confidence interval.you can find free public implementations : google Demetra and TSW.Once you have forecasts, you want to statistically test whether these forecasts are accurate enough for trading purpose or not. In my experience, it is difficult to answer.thanks i will check them out. the should rather support tactical/strategic asset allocation than derive trading positions. i would not expect anything really "tradeable" out of it.

QuoteOriginally posted by: rp10029how do you factor in the recalibration of economic figures?what do you mean by "recalibration"? that the construction methods of the figures change? if so - very valid point. i doubt one could do much about it. just hope the data provider can ...

Recalibration = Regular and irregular series revisions, data collection or compilation methodology changes, discontinuities, etc.

QuoteOriginally posted by: KTERecalibration = Regular and irregular series revisions, data collection or compilation methodology changes, discontinuities, etc.i guess one has to have an eye on it. one of the obvious things is changing consumer baskets for inflation, changing reporting standards for corporates and changing definitions of unemployment figures. do not know how we will handel that. too early stage. i think i found a pretty good source for long term data. http://www.bea.doc.gov quick glance shows that they appear to display detailed stuff long term. peace