Should backtests include or exclude dividends?

[DanielBaeumler]

Today, I came across a tweet from Cesar Alvarenz. He is answering the question whether backtest data should be adjusted by dividends or not.

http://alvarezquanttrading.com/blog/to-dividend-adjust-or-not-to-dividend-adjust-that-is-the-question/

He is referring to Premium data. Apparently, PD is offering two different databases, one included and one excluded dividends (I’m only aware of Premium data delivering data excluding dividends). Cesar ran some tests with both databases verifying the difference and the possible impact on applied indicators.

He is finally recommending to include dividends (if possible) if the strategy is longer term focused and based on an universe holding companies paying noticeable dividends (e.g. S&P 500 or ASX100). For MR systems it seems of less relevance.

Any thoughts?

[Nick Radge]

The NDU has always allowed you to test with or without dividends.

Open Amibroker > File > Database settings > Configure > Select “Capital reconstructions, Special Distributions and Normal Cash Dividends”

However, we’ve always taken the conservative route. See below:

---

AmiBroker has a additional field, Aux2, that we used to show your entitlement to dividends if you held at the close on that date (i.e. exdate – 1 business day) for any dividends that are not already incorporated through adjustments into the price data. The amount is adjusted for any corporate actions you have elected to adjust for (e.g. splits etc.).

That’s all a bit complex…. so here’s some examples:

NYSE:GE had an ordinary dividend of $0.24 with an ex-date 20170915. To be entitled to receive the dividend, you must be holding the stock on 20170914.

As far as displaying in Aux2, it depends:
a) If your price adjustment settings are set to “None”, “Capital reconstructions” or “Capital reconstructions and special distributions”, you’ll see 0.24 in the Aux2 field on 20170914; or
b) If your price adjustment settings are set to “Capital reconstructions, special distributions and ordinary cash dividends” you’ll see 0 in the Aux2 field on 20170914 but all data prior to 20170915 will be proprtionally adjusted for this dividend.

We also offer an indicator called NorgateAllDividendsTimeSeries that allows you to plot the dividends received on the day, irrespective of the Price Adjustment mechanism. The dividend here is adjusted in the same way as the price adjustment method, so effectively the “yield” for that dividend is always the same… eg. a $10 stock giving a $0.10 dividend is effectively 1%. If that stock is subsequently adjusted to $9.5431, you’ll see the indicator return $0.095431.

As far as “Total Return” is concerned when using the adjustment method (b) above, it is done on a proportional basis. eg. for GE on exdate 20170915, dividend of $0.24 with a previous close price on 20170914 of $24.26, we arrive at a “dilution factor” of 0.990107 (i.e. (24.26-0.24)/24.26). This factor is appled to all historical prices prior to 20170925. The effects of multiple dividends are that they are multiplied together. eg. The prior dividend for GE was $0.24 on exdate 20170615, with a prior previous close on 20170614 of $28.69. The dilution factor for that dividend is 0.991635. Therefore any prices prior to 20170615 are multiplied by (0.990107 * 0.991635 = 0.981824754945). This process continues for all new dividends added (plus other corporate actions such as splits etc.).

Now, is this actually showing you a correct Total Return that would match your actual profit/loss on your brokerage account. The answer here is “no” – but it’s a very good approximation. Why the difference? The “Total Return” methodology assumes that you receive the dividend on the entitlement date (exdate – 1) and immediately reinvest it in that stock on the close of the entitlement date. In reality, you actually receive the dividend some weeks later and it might be so fractional that you wouldn’t bother to reinvest it back into the stock because of brokerage costs.

What other option do you have?

You could use the adjustment method of “Capital Reconstructions” so that standard splits/consolidations are taken care of, and use a Custom Backtester within AmiBroker to “inject” any entitled dividends (shown in Aux2) into your account. That would be “more accurate” but again doesn’t take into account the payment date. It might be worthwile considering whether your stops need to be adjusted too.

For the purposes of comparing trading systems, it probably doesn’t matter, but there are always tricky “corner of the envelope” situations that might be an issue. One such area is a company that is still listed but in the process of liquidation. Sometimes the company trades at a price lower than the liquidation price, so a dilution factor cannot be calculated. e.g. A $10 stock makes a (special) liquidation payment of $11. Only adjusting for “capital reconstructions” and “injecting” that liquidation payment into your account your total returns would be correct. Any other method of price adjustment would not work. Another example is calls on partly-paid securities. We actually implement these as a “negative” dividend internally and this is shown as a negative figure in the Aux2 field. Proportional adjustments don’t work very well here especially when the price of the partly-paid security being called is low.

[RobGiles]

Nick, I’ve only just read this, so just want to confirm my understanding.

If I have Amibroker set to > File > Database settings > Configure > “Capital reconstructions and special distributions” I will not be seeing any returns pertaining to dividends in my backtest results?

I’m assuming that these can be added manually or one can make an educated guess as to the additional return relating to divi’s depending on the universe you’re testing?

[Nick Radge]

Just add about 2% to your CAGR. You can switch the dividends on if you want, but its best practice to NOT include them.

[Author]

Posts