French-Fama 2x3, 10x10, mid-caps... Help?
French-Fama 2x3, 10x10, mid-caps... Help?
I've been using the French-Fama data library for various projects, but er, it is a tad nebulous.
http://mba.tuck.dartmouth.edu/pages/fac ... brary.html
Thanks to a bit of advice the good Dr Bernstein posted a while ago, I figured out how to use the "6 Portfolios Formed on Size and Book-to-Market (2 x 3)", and infer the SG, SB, SV, LG, LB, LV historical returns. Now why wouldn't this matrix be 3x3 and include mid-caps, I don't know... Here is WB's post, for reference:
viewtopic.php?t=2520&start=500#p1767160
So I went to the 10x10 matrix. Assuming that I could start from each of those deciles, and easily figure out how to assemble them in a 3x3 matrix of sorts, hence fill the gap of mid-caps (MG, MB, MV).
And... I don't get it. Ignoring the -99.99 data points, and averaging the proper 'slice of deciles' doesn't allow me to reconstruct the 2x3 SG or SV numbers, as a case in point. The math must be more complicated than that because of the value-averaging.
Anybody willing to help?
http://mba.tuck.dartmouth.edu/pages/fac ... brary.html
Thanks to a bit of advice the good Dr Bernstein posted a while ago, I figured out how to use the "6 Portfolios Formed on Size and Book-to-Market (2 x 3)", and infer the SG, SB, SV, LG, LB, LV historical returns. Now why wouldn't this matrix be 3x3 and include mid-caps, I don't know... Here is WB's post, for reference:
viewtopic.php?t=2520&start=500#p1767160
So I went to the 10x10 matrix. Assuming that I could start from each of those deciles, and easily figure out how to assemble them in a 3x3 matrix of sorts, hence fill the gap of mid-caps (MG, MB, MV).
And... I don't get it. Ignoring the -99.99 data points, and averaging the proper 'slice of deciles' doesn't allow me to reconstruct the 2x3 SG or SV numbers, as a case in point. The math must be more complicated than that because of the value-averaging.
Anybody willing to help?
Re: French-Fama 2x3, 10x10, mid-caps... Help?
Hm, I know this is a rather technical question, but... no taker?
I can hardly be the 1st one asking myself the question of reconstructing the mid-cap trajectory since 1927...
I can hardly be the 1st one asking myself the question of reconstructing the mid-cap trajectory since 1927...
Re: French-Fama 2x3, 10x10, mid-caps... Help?
Because mid caps are large caps in CRSP methodology. I don't know what you're trying to do but you need to understand the data you are looking at to draw any meaningful conclusions.
Recent thread: viewtopic.php?f=10&t=183006&p=2776705#p2776705
CRSP: http://www.crsp.com/files/Equity-Indexe ... uide_0.pdf
I don't think you're going to come to any grand conclusions. They outperformed in the aughts, but since then they've substantially lagged both small and large/mega. You tend to find these cycles no matter how you slice the market.
Recent thread: viewtopic.php?f=10&t=183006&p=2776705#p2776705
CRSP: http://www.crsp.com/files/Equity-Indexe ... uide_0.pdf
I don't think you're going to come to any grand conclusions. They outperformed in the aughts, but since then they've substantially lagged both small and large/mega. You tend to find these cycles no matter how you slice the market.
Re: French-Fama 2x3, 10x10, mid-caps... Help?
Thanks for your answer, duncap. This was an interesting post you linked to, I didn't know the CRSP definitions (which are a tad surprising to me).
French-Fama defines things in a somewhat different manner though. As far as I understand... For a given year, they simply divide the world of securities by size and by corresponding deciles. Then the bottom 30% is small-caps, the middle 40% is mid-caps, the top 30% is large-caps. For value/blend/growth, they do something similar based on book-to-price ratios. Then small-value is the intersection of small and value, of course. Strangely enough, they don't provide the numbers for mid-cap value, blend, growth, and I was trying to infer those from a more detailed 10x10 matrix (10 deciles in each direction) they provide, and having some consistency issues...
This being said, yeah this might be rather theoretical, given the mismatch of such definitions with the reality of mid-cap funds we can invest in...
French-Fama defines things in a somewhat different manner though. As far as I understand... For a given year, they simply divide the world of securities by size and by corresponding deciles. Then the bottom 30% is small-caps, the middle 40% is mid-caps, the top 30% is large-caps. For value/blend/growth, they do something similar based on book-to-price ratios. Then small-value is the intersection of small and value, of course. Strangely enough, they don't provide the numbers for mid-cap value, blend, growth, and I was trying to infer those from a more detailed 10x10 matrix (10 deciles in each direction) they provide, and having some consistency issues...
This being said, yeah this might be rather theoretical, given the mismatch of such definitions with the reality of mid-cap funds we can invest in...
Re: French-Fama 2x3, 10x10, mid-caps... Help?
I personally like midcaps a lot. May be recency bias but VG's (+msci index data) have outperformed since 1992, especially MV. I'm only talking about VG funds comparing SV,MV, and LV. I am *not* talking about the extremes of small and value available at DFA.
You probably know one can get the current SMB/HML for any fund from the source: https://www.portfoliovisualizer.com/factor-analysis
Back in 2008 I got interested in this stuff and created a spreadsheet that takes SMB/HML and using the FF data from that time plots inflation corrected graphs. This is in Excel 2007.
Below is a segment of the Excel sheet. It shows the formula for cell N9. Does this look reasonable? I've not puzzled this stuff out in a long time.
You probably know one can get the current SMB/HML for any fund from the source: https://www.portfoliovisualizer.com/factor-analysis
Back in 2008 I got interested in this stuff and created a spreadsheet that takes SMB/HML and using the FF data from that time plots inflation corrected graphs. This is in Excel 2007.
Below is a segment of the Excel sheet. It shows the formula for cell N9. Does this look reasonable? I've not puzzled this stuff out in a long time.
- ClosetIndexer
- Posts: 288
- Joined: Mon Mar 19, 2012 11:00 pm
- Contact:
Re: French-Fama 2x3, 10x10, mid-caps... Help?
What are you trying to accomplish exactly? If you want to use the FF factor data, it would really make sense to learn how to do basic regressions. camontgo's screencast and my basic script here would be a good start: https://www.bogleheads.org/wiki/Fama-Fr ... analysis#R
First though, make sure you understand the 3-factor model itself (which will also allow you to easily understand more recent variations. Something like this might be a good primer: http://papers.ssrn.com/sol3/papers.cfm? ... _id=481881
First though, make sure you understand the 3-factor model itself (which will also allow you to easily understand more recent variations. Something like this might be a good primer: http://papers.ssrn.com/sol3/papers.cfm? ... _id=481881
Re: French-Fama 2x3, 10x10, mid-caps... Help?
What you are doing by averaging is creating an equal weighting of deciles. i.e. the average top 30% of deciles by B/M (that you construct from the 10x10 series) is a different from a cap weighted portfolio of the top 30% by B/M (from the 2x3 series). There is no way to replicate one from the other. Similarly with trying to create a mid-cap series from 10x10 - by averaging it will be an equal weighted decile portfolio, not cap weighted. You could try to create a midcap series from the "Portfolios formed by size" file - weighting each decile return by it market cap e.g. (number of firms x average firm size per decile) * decile returns. Midcap would be about decile 2-5 cap weighted (if I recall). But no way to construct a similar midcap value and growth series.siamond wrote: I don't get it. Ignoring the -99.99 data points, and averaging the proper 'slice of deciles' doesn't allow me to reconstruct the 2x3 SG or SV numbers, as a case in point.
FWIW - here are some annualized returns I have on mid-caps from an earlier Ibbotson Yearbook
1969-2005
Largecap Growth = 9.1%
Largecap Value = 10.8%
Midcap Growth = 9.7%
Midcap Value = 14.0%
Smallcap Growth = 9.2%
Smallcap Value = 15.3%
All Growth = 9.2%
All Value = 11.5%
.
Re: French-Fama 2x3, 10x10, mid-caps... Help?
Robert, do you think that those relative values for SV, MV, LV are representative of Vanguard funds should they have been in existence for the whole period? Or maybe DFA instead with their higher tilt values?
The last 24 years have been somewhat different for msci + VG funds. The data I have for June 1992 through Jan 2016:
LV = 9.1%
LG = 8.3%
MV = 11.4%
MG = 8.6%
SV = 10.6%
SG = 9.4%
The last 24 years have been somewhat different for msci + VG funds. The data I have for June 1992 through Jan 2016:
LV = 9.1%
LG = 8.3%
MV = 11.4%
MG = 8.6%
SV = 10.6%
SG = 9.4%
Last edited by BlueEars on Sat Jan 30, 2016 9:24 am, edited 3 times in total.
Re: French-Fama 2x3, 10x10, mid-caps... Help?
Yes, this is what I suspected. Thank you for the confirmation. Maybe I should just send an e-mail to Ken French and check if they have a 3x3 file somewhere...Robert T wrote:What you are doing by averaging is creating an equal weighting of deciles. i.e. the average top 30% of deciles by B/M (that you construct from the 10x10 series) is a different from a cap weighted portfolio of the top 30% by B/M (from the 2x3 series). There is no way to replicate one from the other.
Re: French-Fama 2x3, 10x10, mid-caps... Help?
Re: French-Fama 2x3, 10x10, mid-caps... Help?
you using too much abbreviation can you just tell me what are they
you using too much abbreviation can you just tell me what are they
Re: French-Fama 2x3, 10x10, mid-caps... Help?
what is french fama ???siamond wrote:I've been using the French-Fama data library for various projects, but er, it is a tad nebulous.
http://mba.tuck.dartmouth.edu/pages/fac ... brary.html
Thanks to a bit of advice the good Dr Bernstein posted a while ago, I figured out how to use the "6 Portfolios Formed on Size and Book-to-Market (2 x 3)", and infer the SG, SB, SV, LG, LB, LV historical returns. Now why wouldn't this matrix be 3x3 and include mid-caps, I don't know... Here is WB's post, for reference:
viewtopic.php?t=2520&start=500#p1767160
So I went to the 10x10 matrix. Assuming that I could start from each of those deciles, and easily figure out how to assemble them in a 3x3 matrix of sorts, hence fill the gap of mid-caps (MG, MB, MV).
And... I don't get it. Ignoring the -99.99 data points, and averaging the proper 'slice of deciles' doesn't allow me to reconstruct the 2x3 SG or SV numbers, as a case in point. The math must be more complicated than that because of the value-averaging.
Anybody willing to help?
Re: French-Fama 2x3, 10x10, mid-caps... Help?
http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/looking wrote:
what is french fama ???
http://www.chicagobooth.edu/faculty/dir ... ene-f-fama
https://en.wikipedia.org/wiki/Fama%E2%8 ... ctor_model
Also, the link already posted.
Re: French-Fama 2x3, 10x10, mid-caps... Help?
Well, sure, but one needs the data first... My point here is that there is missing data (e.g. the midcaps value/blend/growth series).ClosetIndexer wrote:What are you trying to accomplish exactly? If you want to use the FF factor data, it would really make sense to learn how to do basic regressions.
As a side note, regressions are interesting, but I also find quite fascinating to look at John Bogle's telltale charts like I discussed on this thread: viewtopic.php?f=10&t=183017. This actually was my starting point...
Re: French-Fama 2x3, 10x10, mid-caps... Help?
Ah, thank you, I totally forgot that Ibbotson assembled their own growth and value series (starting 1969), and that it included mid-caps indeed. Amusingly, I have the same Ibbotson edition as yours (2006), and just re-read chapter 8.Robert T wrote:FWIW - here are some annualized returns I have on mid-caps from an earlier Ibbotson Yearbook [...]
This actually mostly solves what I was trying to do (get pre-1985 mid-cap value data to get a longer historical perspective than Simba's data provides). 1927 as a starting point would of course be better than 1969, but 1969 is already good progress. Will scan the numbers, draw some charts, and come back with results... Thanks again.
Re: French-Fama 2x3, 10x10, mid-caps... Help?
Ok, I entered the SBBI series in my 'telltale' Excel spreadsheet, and played around... 1969/70/71 were relatively quiet years, so I started from 1972, to be able to more easily compare to other data sources starting in 1972.
See the telltale chart about the value factor (SCV, MCV, LCV) here: viewtopic.php?f=10&t=183017&p=2782080#p2782080
Thanks again to Robert T for reminding me about the SBBI data. I still would like to see the French-Fama version of it (the missing 3x3 matrix), but this is good progress.
See the telltale chart about the value factor (SCV, MCV, LCV) here: viewtopic.php?f=10&t=183017&p=2782080#p2782080
Thanks again to Robert T for reminding me about the SBBI data. I still would like to see the French-Fama version of it (the missing 3x3 matrix), but this is good progress.
Re: French-Fama 2x3, 10x10, mid-caps... Help?
I'm disappointed because I received no comment on the msci + Vanguard fund data that I posted above. No opinions at all? Come on guys.
To be more precise about the data, the msci index was used by Vanguard from as I recall about 2002 to a few years ago when it was replaced by CRSP based indexes. So using the msci index to March 2009 and then the VG data for the asset classes seems reasonable for perspective. I find that the order pretty consistently is MV > SV > LV.
So I have a lot of real life $'s riding on these observations. If I'm wrong then I'd like to hear from people.
To be more precise about the data, the msci index was used by Vanguard from as I recall about 2002 to a few years ago when it was replaced by CRSP based indexes. So using the msci index to March 2009 and then the VG data for the asset classes seems reasonable for perspective. I find that the order pretty consistently is MV > SV > LV.
So I have a lot of real life $'s riding on these observations. If I'm wrong then I'd like to hear from people.
Re: French-Fama 2x3, 10x10, mid-caps... Help?
Here are the years for which I have overlapping data for midcaps.BlueEars wrote:Robert, do you think that those relative values for SV, MV, LV are representative of Vanguard funds should they have been in existence for the whole period? Or maybe DFA instead with their higher tilt values?
1993-2005 Annualized Returns (%)/Standard Deviation
MSCI US MidCap Value = 15.1/15.7
Ibbotson MidCap Value = 14.7/16.2
So in answer to your question - yes, I think the Ibboston MidCap Value series is fairly representative of Vanguard MidCap Value (as you know Vanguard MidCap Value used to track MSCI US Midcap Value.
Re: French-Fama 2x3, 10x10, mid-caps... Help?
You might want to check my entire thread on telltale charts, you should find some useful information and perspective. Value series go through big swings, which are decades long. SV historically outperformed everything else *on the long run*.BlueEars wrote:I'm disappointed because I received no comment on the msci + Vanguard fund data that I posted above. No opinions at all? Come on guys.
To be more precise about the data, the msci index was used by Vanguard from as I recall about 2002 to a few years ago when it was replaced by CRSP based indexes. So using the msci index to March 2009 and then the VG data for the asset classes seems reasonable for perspective. I find that the order pretty consistently is MV > SV > LV.
So I have a lot of real life $'s riding on these observations. If I'm wrong then I'd like to hear from people.
It is a little hard to say about MV because of the lack of old historical data (which is why I opened this thread!), but I can tell you that my somewhat clunky data series derived from the FF 10x10 matrix clearly shows that the consistent order is SV > MV >> LV. This being said, MV is actually close to SV (and was nearly on par in recent times). Again, this is French-Fama I am speaking of here, not any actual fund (Vanguard or DFA).
Re: French-Fama 2x3, 10x10, mid-caps... Help?
I thought more about your point, and tweaked a bit the Simba spreadsheet to compare the Russell+MSCI+Vanguard trajectory with the French-Fama trajectory, the whole thing being calibrated by the TSM trajectory. To be fair, I applied the Vanguard ER to the index data, as typically done in the Simba math. The FF SV trajectory comes straight from the regular FF 2x3 matrix, while I assembled a somewhat approximate FF MV series from the 10x10 matrix, so take the latter with a grain of salt.siamond wrote:BlueEars wrote:I'm disappointed because I received no comment on the msci + Vanguard fund data that I posted above. No opinions at all?
Here is the usual returns vs std-deviation chart:
And here is the telltale chart against TSM (remember, this tracks relative/cumulative growth):
Don't know about you guys, but I can't say that I really like what I see here, notably for Vanguard SCV. While Vanguard MCV is somewhat closer to the FF benchmark. This being said, one could argue that:
1. it might be more an index problem than a Vanguard problem (Vanguard did ok tracking its index) - hopefully CRSP will be better
2. this issue is mostly concentrated in the early years of Vanguard funds - the past 12 years were better aligned to the FF benchmark (on average)
3. this being said, Vanguard (and its index) faltered when it counted the most (i.e. when value grew compared to TSM)
4. MCV looks pretty sweet overall, although I am a tad wary about recency bias
(EDIT: fixed minor issue on charts, more properly applying the Vanguard ER of 0.23% to the FF trajectories)
Last edited by siamond on Mon Feb 01, 2016 9:25 pm, edited 1 time in total.
Re: French-Fama 2x3, 10x10, mid-caps... Help?
Thanks Siamond and Robert.
I'm still digesting that thread on Telltale Charts. I now appreciate how that chart format gives a good perspective. I'm retired and am not thinking in 40 year time frames. So mentally applying a 10 year or 20 year window to these charts is a good idea, I think. What I mean is that if you are, say 65, do you really have to worry about equity returns past age 85? Well maybe yes and maybe no. Depends on your lifespan and your AA as you age.
Siamond, I think the FF series should have a much higher ER then Vanguard. We keep talking about FF but there is no way to get it short of the ER associated with an advisor (+ DFA). Since I'm just going to use non-advisor funds (probably 100% VG) I always need to apply some (unknown) factor to the more extreme FF small and value tilted results. Seems the FF info just makes things confusing.
I'm still digesting that thread on Telltale Charts. I now appreciate how that chart format gives a good perspective. I'm retired and am not thinking in 40 year time frames. So mentally applying a 10 year or 20 year window to these charts is a good idea, I think. What I mean is that if you are, say 65, do you really have to worry about equity returns past age 85? Well maybe yes and maybe no. Depends on your lifespan and your AA as you age.
Siamond, I think the FF series should have a much higher ER then Vanguard. We keep talking about FF but there is no way to get it short of the ER associated with an advisor (+ DFA). Since I'm just going to use non-advisor funds (probably 100% VG) I always need to apply some (unknown) factor to the more extreme FF small and value tilted results. Seems the FF info just makes things confusing.
Re: French-Fama 2x3, 10x10, mid-caps... Help?
Well, it depends on your goals, you still should plan for 30 to 40 years ahead (be positive!), plus you may invest for the next generation or a charitable cause/etc. This being said, you're quite right. This is really the challenge with those value tilts, you could very well have 10 to 20 years without any value-added (pun intended!). Or it could rock the boat like in the early 2000s. Who knows.BlueEars wrote:I'm still digesting that thread on Telltale Charts. I now appreciate how that chart format gives a good perspective. I'm retired and am not thinking in 40 year time frames. So mentally applying a 10 year or 20 year window to these charts is a good idea, I think. What I mean is that if you are, say 65, do you really have to worry about equity returns past age 85? Well maybe yes and maybe no. Depends on your lifespan and your AA as you age.
Yes, that's true. I was naively hoping that the indexes followed by Vanguard would be closer to the FF trajectory, but this doesn't seem to be the case. I did post a telltale chart comparing Vanguard with the DFA funds (which, confusingly enough, follow their own index, which is NOT strictly equivalent to the FF data series!) on the other thread.BlueEars wrote:Siamond, I think the FF series should have a much higher ER then Vanguard. We keep talking about FF but there is no way to get it short of the ER associated with an advisor (+ DFA). Since I'm just going to use non-advisor funds (probably 100% VG) I always need to apply some (unknown) factor to the more extreme FF small and value tilted results. Seems the FF info just makes things confusing.
Re: French-Fama 2x3, 10x10, mid-caps... Help?
Out of curiosity, I followed BlueEars' suggestion, and changed the ER for the FF trajectory from 0.23% (Vanguard-equivalent) to 1.23% (adding 1% ala DFA advisor). As previously mentioned, this may not be entirely significant (DFA using yet another way of indexing), but still, this is interesting.
The Vanguard MCV fund then looks pretty good (so far) - while Vanguard SCV still lags due to its lackluster performance in its early days (reminder, it was created in 1999). Now why can't real-world indexes provide the same performance as the academic form of French-Fama, I don't know...
The Vanguard MCV fund then looks pretty good (so far) - while Vanguard SCV still lags due to its lackluster performance in its early days (reminder, it was created in 1999). Now why can't real-world indexes provide the same performance as the academic form of French-Fama, I don't know...
Re: French-Fama 2x3, 10x10, mid-caps... Help?
One issue I have with K French's data is that most "portfolio's" are formed once/year (June or July IIRC), and monthly changes recorded, together with number of firms. Some portfolio's have a higher/faster rate of decline in the number of firms than others. So to some extent there's a degree of survivorship involved. If a portfolio starts the year (July?) with 1000 firms, but loses 50/month then by the following June its down to 400 firms remaining in the portfolio. The July's figure (gain) comprised 1000 firms, the following June's figure comprised 400 firms. But what about all of the proceeds from those firms that exited the criteria between?!
To replicate in practice I guess you'd have to use the proceeds from disposals to buy the remainder stocks in capital weighted amounts ... which is too expensive/impractical in practice. An alternative would be to assume the proceeds remained in cash, or bought into the total market average, and record the weighted gains, 100% portfolio candidates for the first month, perhaps 60% portfolio gain/40% total market gain for the last month of the (July-June) year. Otherwise in some cases, perhaps non dividend payers for instance, those that exited the portfolio may have done well by year end. If a non dividend selection criteria of one stock became a dividend payer the next month after the year (July) start, then by the following June in having resumed paying dividends the share price might have risen, perhaps a lot. In other cases the reverse might hold. As it stands the data doesn't accurately reflect reasonable comparison figures IMO, which voids any further consideration.
Whether if you did apply portfolio/TSM weighted gains to each months raw data using the number of firms figures produced dissimilar outcomes ??? I did start looking at that some time back, but ran out of time and put it onto the backburner ... but that flame went out.
EDIT : Interesting - from a relatively quick test of proportioning each months gain value according to number of firms (assuming the proceeds of those leaving the portfolio were invested in the broader market return), high book to market value weighted yielded a marginally lower return than equal weighted non dividend stocks !!! 12.7% vs 12.9% annualised respectively (since the 1920's) - statistically the same reward. Counting interim volatility however and the ride was more roller-coaster for non div than for high book to market (which some consider to reflect risk (personally I just hang on tightly so that that 'risk' is neutralised, or even use the volatility to add benefit (buy more when low, sell some when high)).
To replicate in practice I guess you'd have to use the proceeds from disposals to buy the remainder stocks in capital weighted amounts ... which is too expensive/impractical in practice. An alternative would be to assume the proceeds remained in cash, or bought into the total market average, and record the weighted gains, 100% portfolio candidates for the first month, perhaps 60% portfolio gain/40% total market gain for the last month of the (July-June) year. Otherwise in some cases, perhaps non dividend payers for instance, those that exited the portfolio may have done well by year end. If a non dividend selection criteria of one stock became a dividend payer the next month after the year (July) start, then by the following June in having resumed paying dividends the share price might have risen, perhaps a lot. In other cases the reverse might hold. As it stands the data doesn't accurately reflect reasonable comparison figures IMO, which voids any further consideration.
Whether if you did apply portfolio/TSM weighted gains to each months raw data using the number of firms figures produced dissimilar outcomes ??? I did start looking at that some time back, but ran out of time and put it onto the backburner ... but that flame went out.
EDIT : Interesting - from a relatively quick test of proportioning each months gain value according to number of firms (assuming the proceeds of those leaving the portfolio were invested in the broader market return), high book to market value weighted yielded a marginally lower return than equal weighted non dividend stocks !!! 12.7% vs 12.9% annualised respectively (since the 1920's) - statistically the same reward. Counting interim volatility however and the ride was more roller-coaster for non div than for high book to market (which some consider to reflect risk (personally I just hang on tightly so that that 'risk' is neutralised, or even use the volatility to add benefit (buy more when low, sell some when high)).
Re: French-Fama 2x3, 10x10, mid-caps... Help?
Hey Clive, interesting point, but I am not sure to entirely follow you. It seems that your point is that a given firm could be classified in a given category (say SB) when the portfolio is formed in June, then it could potentially be classified in another category (say SV) later in the year due to financial changes (decline?), but FF will still count it as SB for the 12 months to come.Clive wrote:One issue I have with K French's data is that most "portfolio's" are formed once/year (June or July IIRC), and monthly changes recorded, together with number of firms. Some portfolio's have a higher/faster rate of decline in the number of firms than others. So to some extent there's a degree of survivorship involved. If a portfolio starts the year (July?) with 1000 firms, but loses 50/month then by the following June its down to 400 firms remaining in the portfolio. The July's figure (gain) comprised 1000 firms, the following June's figure comprised 400 firms. But what about all of the proceeds from those firms that exited the criteria between?
I don't see that firms are 'lost' in this story, overall they all have to be accounted for in one category or another. Granted, it might be more intellectually satisfying to re-assess the portfolio for a given category every month, but as you said, this would be costly in real-life, and this is just an arbitrary choice. I don't see what's wrong with the arbitrary choice of doing the classification only once a year, and see what goes with such portfolio for the 12 months to come. It's just one way of forming an index among others. Please explain again what's the problem with that?
Re: French-Fama 2x3, 10x10, mid-caps... Help?
Is the portfolios monthly change/gain a direct measure of the portfolio set formed at the end of June (or whatever year start date) for each month, using the end of June equal or value weighted amounts, or a ongoing measure of the stocks still matching the criteria each month, selected out of the original end of June set?It seems that your point is that a given firm could be classified in a given category (say SB) when the portfolio is formed in June, then it could potentially be classified in another category (say SV) later in the year due to financial changes (decline?), but FF will still count it as SB for the 12 months to come
The raw data files typically show monthly value weighed and equal weighted values (gains/change) for each category. The yearly calculation appears to be the product of those monthly values between the chosen year start/end dates. The number of firms for each category are also shown for each month. Which implies that from a set of x initially established at the end of June (or whatever year start point), each month that set is inspected to see whether a firm in x still matches the portfolio criteria and if not it is excluded, otherwise included, and then the included sub-set y has either the value or equal weightings at that the start of that month applied in order to determine that months figure for that category (ongoing average indicator). That approach seems to be more in keeping with the actual yearly change figures presented, which tend to be higher/better than the figures yielded from forming a initial end of June portfolio, held as-is for 12 months.
In short if there are 1000 firms at the end of June start date, but 950 firms a month later, do July figures reflect 1000 firms value (and equal) weighted gains, but August figures reflect 950 firms value (and equal) weighted gains? Or does August figures reflect all 1000 firms value (and equal) weighted gains - even though 50 firms no longer met the portfolio selection criteria? In my tests that attempted to compare/validate, the results tended to indicate more towards the former than the latter. A consequence being that actual (real world) gains lagged the 'average' (calculated/presented figures overstated actuals) before costs and taxes even considered. However since then the methods of calculations/reporting have been changed.
Re: French-Fama 2x3, 10x10, mid-caps... Help?
Ahaah, ok, I understand better your thesis now. I had implicitly assumed it's the former, and you think it's the latter. This would be a bit strange indeed.Clive wrote:Is the portfolios monthly change/gain a direct measure of the portfolio set formed at the end of June (or whatever year start date) for each month, using the end of June equal or value weighted amounts, or a ongoing measure of the stocks still matching the criteria each month, selected out of the original end of June set?
Actually, I don't see the number of firms for each category in the data files. I checked "6_Portfolios_2x3", "25_Portfolios_5x5", "Portfolios_Formed_on_ME", "Portfolios_Formed_on_E-P", and no such information. Maybe you noticed it in earlier versions (e.g. pre 2014), and this got dropped?Clive wrote:The number of firms for each category are also shown for each month. Which implies that from a set of x initially established at the end of June (or whatever year start point), each month that set is inspected to see whether a firm in x still matches the portfolio criteria and if not it is excluded, otherwise included [...]
Making the inference that firms are monthly dropped off (excluded) seems a bit of a leap. You could try to ask the question in an e-mail to Ken French in order to clarify, Pr French is quite approachable.
Re: French-Fama 2x3, 10x10, mid-caps... Help?
The number of firms wouldn't be in the multi-files, just the uni files. I see the number of firms in (as a example) Formed on E-P !!?? http://mba.tuck.dartmouth.edu/pages/fac ... -P_CSV.zipActually, I don't see the number of firms for each category in the data files. I checked "6_Portfolios_2x3", "25_Portfolios_5x5", "Portfolios_Formed_on_ME", "Portfolios_Formed_on_E-P", and no such information
On the front page there's
which does infer that the whole portfolio set is included. I guess that the value/equal weighting applicable at the start of the year would also be fixed (buy and hold for year rather than re-weighted/re-balanced monthly).We have revised the method for computing daily portfolio returns to match more closely the method for computing monthly portfolio returns. Daily files produced before May 2015 keep stocks in a portfolio until the next time it is reconstituted, at the end of June, regardless of the CRSP delist date. Also, a stock is excluded if its price is missing for more than 10 consecutive trading days. Daily files produced after May 2015 drop stocks from a portfolio (i) immediately after their CRSP delist date and (ii) during any period in which they are missing prices for more than 200 consecutive trading days.
Re: French-Fama 2x3, 10x10, mid-caps... Help?
Ah yes, found it, I wasn't looking at the right place. Never noticed this table, took quite some scrolling to locate it. Thank you, I learned something today...Clive wrote:The number of firms wouldn't be in the multi-files, just the uni files. I see the number of firms in (as a example) Formed on E-P !!?? http://mba.tuck.dartmouth.edu/pages/fac ... -P_CSV.zipActually, I don't see the number of firms for each category in the data files. I checked "6_Portfolios_2x3", "25_Portfolios_5x5", "Portfolios_Formed_on_ME", "Portfolios_Formed_on_E-P", and no such information