Inspired by this post https://www.quantopian.com/posts/piotroski-score-plus-aroon-indicator
I split up all of the subcomponents of the Beneish score into individual CustomFactors and then computed the 5 factors only version.
This solved my Out-of-Memory issue! Anyway I hope the the Quantopian Team will give "more power" to the backtesting servers :-)
Here an extract of the code:
quarter_lenght = 65
latest = -1
one_year_ago = -4*quarter_lenght
ttm = [ -1, -quarter_lenght, -2*quarter_lenght, -3*quarter_lenght]
ttm_py = [-4*quarter_lenght, -5*quarter_lenght, -6*quarter_lenght, -7*quarter_lenght]
class DSRI(CustomFactor):
"""
1. DSRI = Days Sales in Receivables Index = (Receivables_t / Revenue_t) / (Receivables_t-1 / Revenue_t-1)
receivable_turnover = revenue / receivables
"""
inputs = [morningstar.operation_ratios.receivable_turnover]
window_length = 4*quarter_lenght + 1
def compute(self, today, assets, out, receivable_turnover):
dsri = receivable_turnover[one_year_ago] / receivable_turnover[latest]
out[:] = dsri
class GMI(CustomFactor):
"""
2. GMI = Gross Margin Index = GrossMargin_t-1 / GrossMargin_t
"""
inputs = [morningstar.operation_ratios.gross_margin]
window_length = 4*quarter_lenght + 1
def compute(self, today, assets, out, gross_margin):
gmi = gross_margin[one_year_ago] / gross_margin[latest]
out[:] = gmi
class AQI(CustomFactor):
"""
3. AQI = Asset Quality Index
AQI = (1 - (CurrentAssets_t + PPE_t) / TotalAssets_t) / (1 - (CurrentAssets_t-1 + PPE_t-1) / TotalAssets_t-1)
"""
inputs = [morningstar.balance_sheet.current_assets,
morningstar.balance_sheet.net_ppe,
morningstar.balance_sheet.total_assets]
window_length = 4*quarter_lenght + 1
def compute(self, today, assets, out, current_assets, net_ppe, total_assets):
aqi = (1 - (current_assets[latest] + net_ppe[latest]) / total_assets[latest])
aqi = aqi / (1 - (current_assets[one_year_ago] + net_ppe[one_year_ago]) / total_assets[one_year_ago])
out[:] = aqi
class SGI(CustomFactor):
"""
4. SGI = Sales Growth Index = Sales_t / Sales_t-1
"""
inputs = [morningstar.operation_ratios.revenue_growth]
window_length = 1
def compute(self, today, assets, out, revenue_growth):
sgi = revenue_growth[latest] + 1.0
out[:] = sgi
class DEPI(CustomFactor):
"""
5. DEPI = Depreciation Index
DEPI = (Depreciation_t-1 / (Depreciaton_t-1 + PPE_t-1)) / (Depreciation_t / (Depreciaton_t + PPE_t))
"""
inputs = [morningstar.balance_sheet.net_ppe,
morningstar.income_statement.depreciation_amortization_depletion]
window_length = 7*quarter_lenght + 1
def compute(self, today, assets, out, net_ppe, depreciation):
dep_ttm = np.sum(depreciation[ttm], axis=0)
dep_ttm_py = np.sum(depreciation[ttm_py], axis=0)
depi = (dep_ttm_py / (dep_ttm_py + net_ppe[one_year_ago])) / (dep_ttm / (dep_ttm + net_ppe[latest]))
out[:] = depi
def initialize(context):
context.count = 0
pipe = Pipeline()
attach_pipeline(pipe, 'my_pipeline')
# 5 Variable Version of the Beneish Model
dsri = DSRI()
gmi = GMI()
aqi = AQI()
sgi = SGI()
depi = DEPI()
mScore = -6.065 + 0.823*dsri + 0.906*gmi + 0.593*aqi + 0.717*sgi + 0.107*depi
pipe.add(mScore, 'mScore')
