Backtesting Crypto in RealTest: Binance Data, Fees and Funding

RealTest has no crypto data source. There is no exchange to select in the import dialog, and there will not be one. What it has instead is a CSV importer, which turns out to be enough, because every crypto exchange will hand you history as CSV and the CSV path in RealTest is a first-class citizen rather than a fallback.

There are two routes in, and the right one depends on whether you are testing a handful of majors or an actual universe.

Route one, Yahoo, for the majors

Yahoo carries the large coins under ticker names like BTC-USD, ETH-USD and SOL-USD, and RealTest imports them the same way it imports a stock.

Import:
	DataSource:	Yahoo
	IncludeList:	SPY
	IncludeList:	BTC-USD, ETH-USD, SOL-USD {"InCRYPTO"}
	StartDate:	Earliest
	EndDate:	Latest
	SaveAs:	MainCrypto.rtd

The braces create a named list, so InList("InCRYPTO") is true for those three symbols and you can keep an equity benchmark like SPY in the same file without it entering trades. This route costs nothing and takes a minute, and for a system that trades Bitcoin and Ethereum it is genuinely sufficient. Its limits are the same ones that apply to Yahoo data anywhere, which is a small set of instruments and no way to reach the ones that died.

Route two, exchange CSV, for everything else

For a universe you pull history from the exchange yourself, usually through CCXT, and write it to CSV. RealTest reads either one file per symbol or a single file containing many symbols, and for crypto the single-file form is easier to keep current.

Import:
	DataSource:	CSV
	CSVFields:	Date,Symbol,Open,High,Low,Close,Volume
	CSVFile:	C:\CryptoData\binance_daily.csv
	StartDate:	Earliest
	EndDate:	Latest
	SaveAs:	crypto.rtd

The file needs a Symbol column when you use CSVFile, and the column names in CSVFields have to match the order in your file. If you instead keep one file per instrument, swap CSVFile for DataPath pointing at the folder, and name each file after its symbol. Both forms accept the same field names, of which only Date and Close are strictly required.

Two details save an afternoon. Dates are parsed as month-day-year unless you set CSVDateFmt: DMY, and exchange exports are frequently day first. And crypto symbols contain a slash, so a comparison in a formula looks like Symbol = $BTC/USDT rather than the bare ticker you would use for a stock.

Fees are percentages, so model them as percentages

Stock backtests use a dollar commission with a minimum. Crypto exchanges charge a fraction of notional and split it by whether you took liquidity or provided it. RealTest expresses that directly, because Commission is a formula rather than a number, and FillValue is the dollar size of the position at entry.

Parameters:
	MakerFee:	0.0002	// limit orders that rest on the book
	TakerFee:	0.0005	// market orders that cross the spread

Strategy:	CryptoMR
	Commission:	TakerFee * FillValue

Set those two values to your own fee tier rather than the example, and use the taker rate unless your entries genuinely rest as limit orders. Getting this wrong is not a rounding error in a strategy that turns over daily. A system taking two hundred round trips a year at market pays a fifth of a percent every time it moves, which compounds into a difference that decides whether the edge exists.

Funding rates, which most backtests just ignore

If you are testing perpetual futures rather than spot, funding is a real cash flow that arrives several times a day and can be larger than your edge in a crowded trade. It is not in the OHLCV file, so it has to come in separately, and the way RealTest handles this is neat. Import the funding history as its own symbol in a second CSV, then read it from any formula with Extern.

Import:
	DataSource:	CSV
	CSVFields:	Date,Symbol,Open,High,Low,Close,Volume
	CSVFile:	C:\CryptoData\binance_daily.csv

	DataSource:	CSV
	CSVFields:	Date,Symbol,Close
	CSVFile:	C:\CryptoData\daily_funding_rates.csv
	SaveAs:	crypto.rtd

Data:
	FundingRate:	Extern($FUNDING_RATE, C)

Multiple DataSource blocks combine into one data file, so the funding series ends up alongside the price series and Extern pulls its value on each bar regardless of which instrument the formula is currently evaluating. From there you can charge it against a position, filter entries when funding is extreme, or simply plot it to see how much of a backtested return was really a funding subsidy.

What crypto breaks that stocks do not

The day has no natural end

Equities have a closing auction. Crypto does not, so a daily bar is a convention someone chose, almost always the UTC midnight boundary. That is fine as long as every file in your universe uses the same convention. Mixing an exchange export cut at UTC with a source cut at New York time produces bars that disagree by hours, which shows up as an edge that appears and disappears depending on which symbols are in the test.

Dead coins are the norm, not the exception

Survivorship bias is worse here than in equities. Exchanges delist pairs regularly, and a token that stops trading simply stops appearing in the export. Build a universe from the pairs Binance lists today and backtest it over five years and you have selected for the tokens that survived a period in which most did not. The fix is the same as it is for stocks, which is to keep the delisted pairs in your CSV rather than rebuilding the list each time you download. If you are new to why that matters, it is the second of the five backtesting mistakes that fake an edge.

There is no consolidated tape

A stock has one official closing price. A coin has as many prices as there are venues, and they differ enough to matter for a strategy trading small moves. Whatever exchange you intend to trade on is the exchange your history should come from, and testing on Binance data while executing on Coinbase introduces a gap that no cost assumption models.

Spot and perpetual are different instruments

They track each other and they are not the same thing. Perpetuals carry funding, allow leverage, and can dislocate from spot exactly when your strategy is most exposed. A backtest built on spot candles that is then traded on perps is testing something adjacent to what you are doing.

How to start

Pull one symbol first, BTC/USDT daily, and import it alone. Check the first and last bar and the row count before writing a strategy, since a truncated download looks identical to a working one until the results make no sense. Then add the rest of the universe, and only after that add funding.

Keep the download and the import as separate steps you can run independently. The download is slow and rate-limited, the import takes seconds, and mixing them means waiting for the exchange every time you want to change a field mapping.

If you would rather not write the downloader, our Python code to download crypto data from Binance with CCXT produces files in the shape the import above expects, and the free crypto CSV data for spot and futures lets you skip the download entirely for a first test.

Once the data is in, the question becomes what to test. If you would rather read than download, the SetupAlpha Substack is where the longer research goes, including the academic papers behind several of these systems. It is free and the archive is large enough to keep you busy.