The issue of dividend adjustment.

[sokurm]

Hello.

I've searched this forum and other places on the ent on the issue of dividend adjustment in stocks data. What I've gathered so far is :
* dividend-adjustment is good, because it gives you a better return estimate when back-Dtesting. Dividend is part of the return and should not be ignored.
* Dividend-adjustment is bad because they somehow distort the chart. Support and resistance levels, for example are usually drawn on non-adjusted prices.

This contradiction, of course, is problematic. The most sensible thing to do when back-testing therefore seems to :
* generate buy/sell orders on unadjusted data.
* calculate return by transposing those buy/sell orders on dividend-adjusted data.

So my question is this :

Is there a combination of software/data feed that will easily allow me to do so. I'm basically thinking about
* unadjusted data, but with the dividend information supplied so I cam make my own adjustment for return analysis
* a program that would not make the whole process into a chore.

Does anyone have experience with this issue, or simply food for thought to add to this debate?

I'd be most grateful.
Sokurm.

 

[kipptrader]

For short/medium term testing, most dividends can ignored as they have a negligible (if any) effect on results.

I say most dividends because some do have a real effect on price. These 'special' dividends are reflected in the trading price the next session after payout. These are always adjusted for in the price.

From wiki: "A prominent example of a special dividend was the $3 dividend announced by Microsoft in 2004 to partially relieve its balance sheet of a large cash balance." See yahoo history.

If you are doing very long term back-testing (many decades), unadjusted data even from stock splits is preferred so you don't have very low volatility and small ranges in the price data.

 

[RichardDale]

For short/medium term testing, most dividends can ignored as they have a negligible (if any) effect on results.

I say most dividends because some do have a real effect on price. These 'special' dividends are reflected in the trading price the next session after payout. These are always adjusted for in the price.

The standard methodology employed by Dow Jones, Standard & Poors, Reuters and Blomberg always adjusts for special dividends plus all of the normal capital adjustments (splits, reverse splits, rights issues, capital returns, demergers/spinoffs). They do not adjusted for normal cash dividends except in Total Return charts, so that assessment appears sound. However, many US-listed stocks have comparatively little or no dividend payouts anyway - just look at most of the NASDAQ-listed companies' dividend policies.

I disagree with the following by Kipptrader:

f you are doing very long term back-testing (many decades), unadjusted data even from stock splits is preferred so you don't have very low volatility and small ranges in the price data.

If you do not adjust for stock splits etc. then the volatilty will be vastly distorted and significantly increased.

If it is adjusted correction (proportioanlly) then the scale of volaility of the historical data is maintained.

For what it's worth we're looking to enhance our data products for suitable backtesting software to allow easy reconciliation of dividend returns, perform standard adjustments (as described above) and also expose the original closing price. It's a big job though, typically only undertaken by corporate/institutional clients. But at least we have the data to do it.

 

[kipptrader]

If it is adjusted correction (proportioanlly) then the scale of volaility of the historical data is maintained.

I agree. But most data vendors do just scale the splits. For a company with a large number of splits over its history, you end up with a range of 0.01 or 0.02 and a price value of 0.10 for a large duration of prices in the distant past.

I should have also mentioned if you are using unadjusted data, you need back-testing software that will adjust for splits as you walf-forward through the data.

 

[RichardDale]

I'll give an example... take General Electric.

In its history, it's had:
May 2000: 3:1 Stock split (factor 0.33333333)
May 1997 2:1 stock split (factor 0.5)
May 1994: 2:1 stock split (factor 0.5)
May 1987: 2:1 stock split (factor 0.5)
June 1983: 2:1 stock split (factor 0.5)
June 1971: 2:1 stock split (factor 0.5)
June 1954: : 3:1 stock split (factor 0.33333333)

Multiplying all of these together gives a cumulative dilution factor of 0.0034722222187465

If we have a look at the prices from, January 1953

Date,OHLCV
2 Jan 1953: 72.875,72.875,72.125,72.5,3000
5 Jan 1953: 72.5,73.5,72.375,73.25,6100

Note that the ranges on 2 Jan 1953 is 0.75. As a percentage of the close price it's about 1%

If we perform the adjustment this now shows:
2 Jan 1953: 0.24999869661328,0.24999869661328,0.247425811227894,0.248712253920587,961955.01519757
5 Jan 1953: 0.248712253920587,0.252142767767767,0.248283439689689,0.251285139305972,1778159.27051672

The HL range is still 1% of the closing price, and the volume is consistent too, which allows your position sizing algorithms to work. Other forms of volatility (eg. deviations, ATR%, historical volalitility) all work out to the same percentage value.

The problem that I think you're aluding to is when you have data vendors that round the number of decimal places to 2 for historical data. This is silly and definitely does distort historical volatility.

By having the data adjusting correctly you do not need to worry about the effect of splits etc. in your backtesting. Even better, you do not need backtesting software to handle this for you. Very few backtesting software packages out there handle corporate action events anyway.

 

[ceyhantuzcu]

Hello Richard,
Do you know how we can read SPLIT FACTOR from yahoo finance servers?
For instance;
http://finance.yahoo.com/d/quotes.csv?s=AAPL&f=snd1l1yr
this links give some financial values of AAPL
like as s= stock code, n= stock name
I look for tag to read SPLIT FACTOR bur still coouln't find it.

Thanks,

I'll give an example... take General Electric.

In its history, it's had:
May 2000: 3:1 Stock split (factor 0.33333333)
May 1997 2:1 stock split (factor 0.5)
May 1994: 2:1 stock split (factor 0.5)
May 1987: 2:1 stock split (factor 0.5)
June 1983: 2:1 stock split (factor 0.5)
June 1971: 2:1 stock split (factor 0.5)
June 1954: : 3:1 stock split (factor 0.33333333)

Multiplying all of these together gives a cumulative dilution factor of 0.0034722222187465

If we have a look at the prices from, January 1953

Date,OHLCV
2 Jan 1953: 72.875,72.875,72.125,72.5,3000
5 Jan 1953: 72.5,73.5,72.375,73.25,6100

Note that the ranges on 2 Jan 1953 is 0.75. As a percentage of the close price it's about 1%

If we perform the adjustment this now shows:
2 Jan 1953: 0.24999869661328,0.24999869661328,0.247425811227894,0.248712253920587,961955.01519757
5 Jan 1953: 0.248712253920587,0.252142767767767,0.248283439689689,0.251285139305972,1778159.27051672

The HL range is still 1% of the closing price, and the volume is consistent too, which allows your position sizing algorithms to work. Other forms of volatility (eg. deviations, ATR%, historical volalitility) all work out to the same percentage value.

The problem that I think you're aluding to is when you have data vendors that round the number of decimal places to 2 for historical data. This is silly and definitely does distort historical volatility.

By having the data adjusting correctly you do not need to worry about the effect of splits etc. in your backtesting. Even better, you do not need backtesting software to handle this for you. Very few backtesting software packages out there handle corporate action events anyway.

 

[RichardDale]

Yahoo does not supply dilution factors (aka split factors).

Different Dilution factors are applicable at different points in the stock's history - a single dilution factor does not exist for the history of a stock.

 

[ceyhantuzcu]

Thanks Richard,
I was just in contact with Yahoo Finance about the issue and they confirmed your saying.
They have not such a data providing.
And i now, just fill out the "feedback" form to Yahoo. Maybe they decide to do...
Anyway, thanks again for your reply.

Yahoo does not supply dilution factors (aka split factors).

Different Dilution factors are applicable at different points in the stock's history - a single dilution factor does not exist for the history of a stock.