In financial econometrics research, it is very common to investigate relationships between financial time series that take the form of daily data. The variable will often be made $I(0)$ by taking the log difference, for example; $\ln(P_t)-\ln(P_{t-1})$.

However, daily data means that there's $5$ data points each week, and Saturday and Sunday are missing. This seems to get no mention in the applied literature that I'm aware of. Here's some closely related questions that I have that come from this observation:

  • Does this qualify as irregularly spaced data, even though financial markets are closed over the weekend?

  • If so, what are the consequences for the validity of extant empirical results garnered thus far in the gigantic number of papers that ignore this issue?

  • 6
    $\begingroup$ Regarding your first question, this problem is sometimes called weekend effect. In my opinion, the answer is context-dependent. For instance, this question makes a lot of sense in the case of stock returns. See for instance here, here, here and here. But I am not sure if this effect applies to other contexts. $\endgroup$ – user10525 Dec 15 '12 at 15:52
  • $\begingroup$ @Procrastinator Submit answer it's very good!! $\endgroup$ – Jase Dec 15 '12 at 16:19
  • $\begingroup$ There is a quantitative finance SE that may be more suited to get meaningfull answers. There are actualy a lot more problems than weekends: nights, bank holidays... etc. which get worse with multiple price sources. $\endgroup$ – lcrmorin Jul 22 '19 at 9:41

Full disclosure! I don't know about finance/economy, so sorry in advance for my ignorance. But I find this question wider than finance. Analyzing irregularly sampled data arises in many other fields, such as biology and medicine. One of the shortcomings of classical approaches like Autoregressive Regression (AR) is their weakness in dealing with irregularly sampled data. However this problem can be tackled by Gaussian processes (GPs). It's used for example here or here.

| cite | improve this answer | |

Traditionally, we don't worry about non trading days and count this as regularly spaced data. There are however two possible effects that you'd have to worry about.

The first is the effect of time on momentum and interaction with leading indicators. If you have a lagged variable that is a good leader - let's say it's mean temperature - then some of your data points will be lagged to the next day (Friday -> Thurs) while others are lagged three days (Monday -> Friday). There's likely to be spurious results because of that.

The second issue is activity that happens when markets are closed. After hours trading, options pricing, etc. If those are a factor, you may be better off calculating a regularly spaced time series and interpolating or accounting for non-trading days some other way.

| cite | improve this answer | |
  • $\begingroup$ Just because markets are closed doesn't mean it's regularly spaced. If we think of it as an underlying process that we sample discretely (when markets are opened) but still evolves when markets are closed then it is irregular. I think this continual evolution metaphor is more useful since it is consistent with close to open jumps (all information from closed times being revealed in 1 moment). $\endgroup$ – Jase Oct 31 '13 at 16:56

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.