This package is a proof of concept of TimeTable interface, which is a small extension of Tables.jl. This project should be used for demonstration purposes only, since it contains heavy pirating of TimeSeries.jl, Temporal.jl and Timestamps.jl.
Since methods of different packages are pirated, internal versions should be used. Messages like "recompilation can be broken" should be ignored, since it is an expected behaviour. Ideally these methods should land in respective packages and no messages of this kind appear.
Since package is not registered it should be installed with
julia> using Pkg; Pkg.dev("https://github.com/Arkoniak/ProtoMarketData.jl")To prepare workspace do
using ProtoMarketData
using Indicators
using MarketTechnicals
const TSPArray = ProtoMarketData.Timestamps.TimestampArray
const DataFrame = ProtoMarketData.DataFrames.DataFrame
const TimeArray = ProtoMarketData.TimeSeries.TimeArray
const TS = ProtoMarketData.Temporal.TSIn this approach, all data sources emit special WrapperTimeTable which has istimetable trait. As TimeTable it contains information regarding time axis and all other interfaces are modified to respect timeaxis information
julia> raw = yahoo("SPY")
ProtoMarketData.WrapperTimeTable{CSV.File{true}}(:Date, CSV.File("<raw buffer>"):
Size: 7098 x 7
Tables.Schema:
:Date Date
:Open Float64
:High Float64
:Low Float64
:Close Float64
Symbol("Adj Close") Float64
:Volume Int64)This wrapper object can be transformed to any TimeTable types
julia> tsp1 = TSPArray(raw)
7098×6 Timestamps.TimestampArray{Date, Tuple{Float64, Float64, Float64, Float64, Float64, Int64
}}:
1993-01-29 | (43.96875, 43.96875, 43.75, 43.9375, 25.884184, 1003200)
1993-02-01 | (43.96875, 44.25, 43.96875, 44.25, 26.068277, 480500)
1993-02-02 | (44.21875, 44.375, 44.125, 44.34375, 26.123499, 201300)
1993-02-03 | (44.40625, 44.84375, 44.375, 44.8125, 26.399649, 529400)
⋮
julia> ta1 = TimeArray(raw)
7098×6 TimeSeries.TimeArray{Float64, 2, Date, Matrix{Float64}} 1993-01-29 to 2021-04-07
│ │ Open │ High │ Low │ Close │ Adj Close │ Volume │
├────────────┼─────────┼─────────┼─────────┼─────────┼───────────┼────────────┤
│ 1993-01-29 │ 43.9688 │ 43.9688 │ 43.75 │ 43.9375 │ 25.8842 │ 1.0032e6 │
│ 1993-02-01 │ 43.9688 │ 44.25 │ 43.9688 │ 44.25 │ 26.0683 │ 480500.0 │
│ 1993-02-02 │ 44.2188 │ 44.375 │ 44.125 │ 44.3438 │ 26.1235 │ 201300.0 │
│ 1993-02-03 │ 44.4062 │ 44.8438 │ 44.375 │ 44.8125 │ 26.3996 │ 529400.0 │
⋮
julia> ts1 = TS(raw)
7098x6 Temporal.TS{Float64, Date}: 1993-01-29 to 2021-04-07
Index Open High Low Close Adj Close Volume
1993-01-29 43.9688 43.9688 43.75 43.9375 25.8842 1.0032e6
1993-02-01 43.9688 44.25 43.9688 44.25 26.0683 480500.0
1993-02-02 44.2188 44.375 44.125 44.3438 26.1235 201300.0
1993-02-03 44.4062 44.8438 44.375 44.8125 26.3996 529400.0
⋮Since this object is also Tables.jl object any other sinks can used it too (of course, time axis information is lost in this case)
julia> df = DataFrame(raw)
7098×7 DataFrame
Row │ Date Open High Low Close Adj Close Volume
│ Date Float64 Float64 Float64 Float64 Float64 Int64
──────┼──────────────────────────────────────────────────────────────────────────
1 │ 1993-01-29 43.9688 43.9688 43.75 43.9375 25.8842 1003200
2 │ 1993-02-01 43.9688 44.25 43.9688 44.25 26.0683 480500
3 │ 1993-02-02 44.2188 44.375 44.125 44.3438 26.1235 201300
4 │ 1993-02-03 44.4062 44.8438 44.375 44.8125 26.3996 529400
⋮Since all TimeTable types support timeaxis data, you can freely transform time tables from one type to another
# From timestamps to TS
julia> tsp1 |> TS
7098x6 Temporal.TS{Float64, Date}: 1993-01-29 to 2021-04-07
Index Open High Low Close Adj Close Volume
1993-01-29 43.9688 43.9688 43.75 43.9375 25.8842 1.0032e6
1993-02-01 43.9688 44.25 43.9688 44.25 26.0683 480500.0
1993-02-02 44.2188 44.375 44.125 44.3438 26.1235 201300.0
1993-02-03 44.4062 44.8438 44.375 44.8125 26.3996 529400.0
⋮
# Further from TS to TimeArray
julia> tsp1 |> TS |> TimeArray
7098×6 TimeSeries.TimeArray{Float64, 2, Date, Matrix{Float64}} 1993-01-29 to 2021-04-07
│ │ Open │ High │ Low │ Close │ Adj Close │ Volume │
├────────────┼─────────┼─────────┼─────────┼─────────┼───────────┼────────────┤
│ 1993-01-29 │ 43.9688 │ 43.9688 │ 43.75 │ 43.9375 │ 25.8842 │ 1.0032e6 │
│ 1993-02-01 │ 43.9688 │ 44.25 │ 43.9688 │ 44.25 │ 26.0683 │ 480500.0 │
│ 1993-02-02 │ 44.2188 │ 44.375 │ 44.125 │ 44.3438 │ 26.1235 │ 201300.0 │
│ 1993-02-03 │ 44.4062 │ 44.8438 │ 44.375 │ 44.8125 │ 26.3996 │ 529400.0 │
⋮
# And back to Timestamps
julia> tsp1 |> TS |> TimeArray |> TSPArray
7098×6 Timestamps.TimestampArray{Date, NTuple{6, Float64}}:
1993-01-29 | (43.96875, 43.96875, 43.75, 43.9375, 25.884184, 1.0032e6)
1993-02-01 | (43.96875, 44.25, 43.96875, 44.25, 26.068277, 480500.0)
1993-02-02 | (44.21875, 44.375, 44.125, 44.34375, 26.123499, 201300.0)
1993-02-03 | (44.40625, 44.84375, 44.375, 44.8125, 26.399649, 529400.0)
⋮Since all transformations support timestamp keyword as a fallback, even non timetable types can be used in transformation chains, yet slightly less convenient
# Convert DataFrame to Timestamps and than to TS
julia> TSPArray(df; timestamp = :Date) |> TS
7098x6 Temporal.TS{Float64, Date}: 1993-01-29 to 2021-04-07
Index Open High Low Close Adj Close Volume
1993-01-29 43.9688 43.9688 43.75 43.9375 25.8842 1.0032e6
1993-02-01 43.9688 44.25 43.9688 44.25 26.0683 480500.0
1993-02-02 44.2188 44.375 44.125 44.3438 26.1235 201300.0
1993-02-03 44.4062 44.8438 44.375 44.8125 26.3996 529400.0
⋮
One immediate benefit of this transparent and seamless conversions is possibility to use any package to make any kind of calculations, if current type has no support for any particular operation. This way calculations can be gathered in one huge package or spread in dozen of small focused packages, user can use them all. Of course, one should be careful and not use conversions in tight loops. Still it covers lot of practical cases.
There is no PSAR indicator at the moment for TimeArray type in MarketTechnicals.jl, but there is one in Indicators.jl for Temporal
julia> ta1 |> TS |> psar |> TimeArray
7098×1 TimeSeries.TimeArray{Float64, 1, Date, Vector{Float64}} 1993-01-29 to 2021-04-07
│ │ PSAR │
├────────────┼──────────┤
│ 1993-01-29 │ 42.0817 │
│ 1993-02-01 │ 42.1195 │
│ 1993-02-02 │ 42.2047 │
│ 1993-02-03 │ 42.3349 │
⋮We can compare different implementations of simple moving average between different libraries
julia> MarketTechnicals.sma(ta1[[:Close]], 3) |> TS
7096x1 Temporal.TS{Float64, Date}: 1993-02-02 to 2021-04-07
Index Close_sma_3
1993-02-02 44.1771
1993-02-03 44.4688
1993-02-04 44.7188
⋮
julia> Indicators.sma(TS(ta1[[:Close]]), n = 3)
7098x1 Temporal.TS{Float64, Date}: 1993-01-29 to 2021-04-07
Index SMA
1993-01-29 NaN
1993-02-01 NaN
1993-02-02 44.1771
1993-02-03 44.4688
1993-02-04 44.7188
⋮We can apply MarketTechnicals.rsi and convert result in TimestampArray for further usage in backtesting
julia> tsp1 |> TimeArray |> x -> x[[:Close]] |> MarketTechnicals.rsi |> TSPArray
7084×1 Timestamps.TimestampArray{Date, Tuple{Float64}}:
1993-02-19 | (44.444444444444436,)
1993-02-22 | (49.4818652849741,)
1993-02-23 | (48.46769471013078,)
1993-02-24 | (63.854279665925915,)
1993-02-25 | (65.81704734556533,)
⋮We can apply Indicators.macd for the same reason
julia> tsp1 |> TS |> Indicators.macd |> TSPArray
7098×3 Timestamps.TimestampArray{Date, Tuple{Float64, Float64, Float64}}:
1993-01-29 | (NaN, NaN, NaN)
1993-02-01 | (NaN, NaN, NaN)
⋮
2021-03-05 | (0.24446773578944203, 1.9126881594256007, -1.6682204236361586)
2021-03-08 | (-0.001533747188830148, 1.5243867936480027, -1.5259205408368328)
2021-03-09 | (0.23908709375064063, 1.162151050415692, -0.9230639566650514)
⋮