Importing and Cleaning Market Data in MATLAB: timetable, retime and synchronize

Importing and Cleaning Market Data in MATLAB: timetable, retime and synchronize

Every backtest and every indicator you compute in MATLAB is only as good as the data underneath it. Yet data import and cleaning is the step most traders rush — they load a CSV, the dates parse as text, two feeds do not line up, a few bars are missing, and the bug surfaces three steps later as a nonsensical equity curve. This guide covers getting market data into MATLAB cleanly and keeping it that way, built around the timetable and its two workhorse functions, retime and synchronize.

Why a timetable, not a matrix

A plain matrix throws away the one thing that makes market data special: time. A timetable carries a proper datetime row index alongside your variables, so every operation that follows knows when each observation happened. That is what makes alignment and resampling reliable instead of error-prone index arithmetic.

Reading a CSV the right way

T = readtimetable("AAPL.csv");

% If the time column is not auto-detected, be explicit:

opts = detectImportOptions("AAPL.csv");

opts = setvartype(opts, "Date", "datetime");

T = readtimetable("AAPL.csv", opts);



T.Properties.DimensionNames{1}   % name of the time dimension

head(T)

The most common import bug is dates read as text. Confirm the index is a real datetime (isdatetime(T.Time)) before doing anything else — everything downstream depends on it.

Cleaning: the four things that bite

• Duplicate timestamps — two rows for the same bar. Remove them deterministically:
  

T = sortrows(T);                 % ensure chronological order first

T = unique(T);                   % drop exact duplicate rows

% or keep the last quote for each timestamp:

[~, idx] = unique(T.Time, "last");

T = T(idx, :);

• Missing values — gaps in the feed. fillmissing gives you control over how:
  

T = fillmissing(T, "previous");  % carry last price forward (sensible for prices)

% T = fillmissing(T, "linear");  % interpolate (rarely right for prices)

• Irregular timestamps — bars that do not land on a clean grid. That is what retime fixes (next section).
  
• Outliers / bad ticks — a zero or a fat-finger spike. rmoutliers or a simple sanity filter catches the obvious ones; never let a single bad print define your high or low.
  

retime: put one series on a regular grid

retime resamples a single timetable onto a regular time vector, aggregating or interpolating as you choose:

% Daily close aggregated to weekly OHLC-style summaries

daily   = retime(T, "daily", "previous");        % regular daily grid

weekly  = retime(T, "weekly", "lastvalue");      % week-ending values

% Custom aggregation per variable:

hourly  = retime(T, "hourly", @mean);            % average within each hour

Choose the method to match the variable: lastvalue/previous for prices, sum for volume, mean for rates.

synchronize: align two or more feeds

When you have AAPL and MSFT — or price and an economic series on a different calendar — synchronize merges them onto a common time base:

both = synchronize(aaplTT, msftTT);                 % union of all times

both = synchronize(aaplTT, msftTT, "union", "previous");  % fill gaps forward

both = synchronize(aaplTT, msftTT, "intersection");       % only shared times

union keeps every timestamp (with fills), intersection keeps only times present in both. Picking the wrong one silently changes your sample — decide which you actually want.

A reusable import-and-clean recipe

function T = loadClean(file)

    T = readtimetable(file);

    T = sortrows(T);

    [~, idx] = unique(T.Time, "last");   % drop duplicate timestamps

    T = T(idx, :);

    T = fillmissing(T, "previous");      % carry prices forward

    T = retime(T, "daily", "previous");  % regular daily grid

end

Wrap your loading in a function like this and every script starts from clean, regular, aligned data — instead of each one re-discovering the same gaps and duplicates.

Habits worth keeping

• Validate the index first: confirm it is datetime, sorted, and free of duplicates before computing anything.
  
• Pick fill and aggregation methods deliberately — "previous" for prices, "sum" for volume; the default is rarely what you want for finance.
  
• Synchronize before you compute any cross-asset signal, or you are silently comparing different days.
  
• Keep the cleaning in one function so a data refresh is reproducible and every downstream result is built on the same foundation.
  

Clean import is unglamorous and it is exactly where most "mysterious" backtest results are born. Get the timetable, the retime grid and the synchronize alignment right at the start, and the rest of your MATLAB work rests on solid ground.

Wrestling with a particular feed — duplicates, gaps, mismatched calendars? Describe it and we can build the retime/synchronize call together.